Download the Full Ebook and Access More Features - ebooknice.

com

(Ebook) Computer Vision Projects with PyTorch:

Design and Develop Production-Grade Models by
Akshay Kulkarni, Adarsha Shivananda, Nitin Ranjan
Sharma ISBN 9781484282724, 1484282728
https://ebooknice.com/product/computer-vision-projects-with-
pytorch-design-and-develop-production-grade-models-44187626

OR CLICK HERE

DOWLOAD EBOOK

Download more ebook instantly today at https://ebooknice.com

 Instant digital products (PDF, ePub, MOBI) ready for you
Download now and discover formats that fit your needs...

Start reading on any device today!

(Ebook) Biota Grow 2C gather 2C cook by Loucas, Jason;

Viles, James ISBN 9781459699816, 9781743365571,
9781925268492, 1459699815, 1743365578, 1925268497
https://ebooknice.com/product/biota-grow-2c-gather-2c-cook-6661374

ebooknice.com

(Ebook) Natural Language Processing Projects: Build Next-

Generation NLP Applications Using AI Techniques by Akshay
Kulkarni, Adarsha Shivananda, Anoosh Kulkarni ISBN
9781484273852, 1484273850
https://ebooknice.com/product/natural-language-processing-projects-
build-next-generation-nlp-applications-using-ai-techniques-36506460

ebooknice.com

(Ebook) Applied Generative AI for Beginners: Practical

Knowledge on Diffusion Models, ChatGPT, and Other LLMs by
Akshay Kulkarni, Adarsha Shivananda, Anoosh Kulkarni,
Dilip Gudivada ISBN 9781484299937, 1484299930
https://ebooknice.com/product/applied-generative-ai-for-beginners-
practical-knowledge-on-diffusion-models-chatgpt-and-other-
llms-53764360
ebooknice.com

(Ebook) Natural Language Processing Recipes: Unlocking

Text Data with Machine Learning and Deep Learning Using
Python by Akshay Kulkarni, Adarsha Shivananda ISBN
9781484273500, 1484273508
https://ebooknice.com/product/natural-language-processing-recipes-
unlocking-text-data-with-machine-learning-and-deep-learning-using-
python-34204404
ebooknice.com
(Ebook) Introduction to Prescriptive AI: A Primer for
Decision Intelligence Solutioning with Python by Akshay
Kulkarni, Adarsha Shivananda, Avinash Manure ISBN
9781484295670, 1484295676
https://ebooknice.com/product/introduction-to-prescriptive-ai-a-
primer-for-decision-intelligence-solutioning-with-python-50765970

ebooknice.com

(Ebook) Time Series Algorithms Recipes: Implement Machine

Learning and Deep Learning Techniques with Python by
Akshay Kulkarni, Adarsha Shivananda, Anoosh Kulkarni, V
Adithya Krishnan ISBN 9781484289785, 9781484289778,
https://ebooknice.com/product/time-series-algorithms-recipes-
1484289781, 1484289773
implement-machine-learning-and-deep-learning-techniques-with-
python-47430754
ebooknice.com

(Ebook) Natural Language Processing Recipes: Unlocking

Text Data with Machine Learning and Deep Learning Using
Python by Akshay Kulkarni, Adarsha Shivananda ISBN
9781484273517, 9781484273500, 1484273508, 1484273516
https://ebooknice.com/product/natural-language-processing-recipes-
unlocking-text-data-with-machine-learning-and-deep-learning-using-
python-34204734
ebooknice.com

(Ebook) Matematik 5000+ Kurs 2c Lärobok by Lena

Alfredsson, Hans Heikne, Sanna Bodemyr ISBN 9789127456600,
9127456609
https://ebooknice.com/product/matematik-5000-kurs-2c-larobok-23848312

ebooknice.com

(Ebook) SAT II Success MATH 1C and 2C 2002 (Peterson's SAT

II Success) by Peterson's ISBN 9780768906677, 0768906679

https://ebooknice.com/product/sat-ii-success-
math-1c-and-2c-2002-peterson-s-sat-ii-success-1722018

ebooknice.com
Akshay Kulkarni, Adarsha Shivananda and Nitin Ranjan Sharma

Computer Vision Projects with PyTorch

Design and Develop Production-Grade Models
Akshay Kulkarni
Bangalore, Karnataka, India

Adarsha Shivananda
Hosanagara tq, Shimoga dt, Karnataka, India

Nitin Ranjan Sharma

Bangalore, India

ISBN 978-1-4842-8272-4 e-ISBN 978-1-4842-8273-1

https://doi.org/10.1007/978-1-4842-8273-1

© Akshay Kulkarni, Adarsha Shivananda, and Nitin Ranjan Sharma

2022

Apress Standard

The use of general descriptive names, registered names, trademarks,

service marks, etc. in this publication does not imply, even in the
absence of a specific statement, that such names are exempt from the
relevant protective laws and regulations and therefore free for general
use.

The publisher, the authors and the editors are safe to assume that the
advice and information in this book are believed to be true and accurate
at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, expressed or implied, with respect to the
material contained herein or for any errors or omissions that may have
been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Apress imprint is published by the registered company APress

Media, LLC, part of Springer Nature.
The registered company address is: 1 New York Plaza, New York, NY
10004, U.S.A.
To our families.
Introduction
This book explores various popular methodologies in the field of
computer vision in order to unravel its mysteries. We use the PyTorch
framework, because it's used by researchers, developers, and beginners
to leverage the power of deep learning. This book explores multiple
computer vision problems and shows you how to solve them. You can
expect an introduction to some of the most critical challenges with
hands-on code in PyTorch, which is suitable for beginner and
intermediate Python users, along with various methodologies used to
solve those business problems.
Production-grade code related to important concepts we present
over the course of the book will help you get started quickly. These code
snippets can be run on local systems, with or without GPUs (Graphics
Processing Units) or on a cloud platform.
We’ll introduce you to the concepts of image processing in stages,
starting with the basic concepts of computer vision in the first chapter.
We’ll also delve into the field of deep learning and explain how models
are developed for vision-related tasks. You’ll get a quick introduction to
PyTorch to prepare you for the example business challenges we’ll be
presenting later in the book. We explore concepts of the revolutionary
convolutional neural networks, as well as architectures such as VGG,
ResNet, YOLO, Inception, R-CNN, and many others.
The book dives deep into business problems related to image
classification, object detection, and segmentation. We explore the
concepts of super-resolution and GAN architectures, which are used in
many industries. You learn about image similarity and pose estimation,
which help with unsupervised problem sets. There are topics related to
video analytics, which will help you develop the mindset of using the
image and time-based concepts of frames. Adding to the list, the book
ends by discussing how these deep learning models can be explained to
your business partners. This book aims to be a complete suite for those
pursuing computer vision business problems.
Any source code or other supplementary material referenced by the
author in this book is available to readers on GitHub via the book’s
product page, located at www.apress.com/978-1-4842-8272-4. For
more detailed information, please visit
http://www.apress.com/source-code.
Table of Contents
Chapter 1:​The Building Blocks of Computer Vision
What Is Computer Vision
Applications
Channels
Convolutional Neural Networks
Understanding the CNN Architecture Type
Working with Deep Learning Model Techniques
Introduction to PyTorch
Summary
Chapter 2:​Image Classification
Topics to Cover
Defining the Problem
Overview of the Approach
Creating an Image Classification Pipeline
First Basic Model
Data
Data Exploration
Data Loader
Define the Model
The Training Process
The Second Variation of Model
The Third Variation of the Model
The Fourth Variation of the Model
Summary
Chapter 3:​Building an Object Detection Model
 Object Detection Using Boosted Cascade
R-CNN
The Region Proposal Network
Fast Region-Based Convolutional Neural Network
How the Region Proposal Network Works
The Anchor Generation Layer
The Region Proposal Layer
Mask R-CNN
Prerequisites
YOLO
YOLO V2/​V3
Project Code Snippets
Step 1:​Getting Annotated Data
Step 2:​Fixing the Configuration File and Training
The Model File
Summary
Chapter 4:​Building an Image Segmentation Model
Image Segmentation
Pretrained Support from PyTorch
Semantic Segmentation
Instance Segmentation
Fine-Tuning the Model
Summary
Chapter 5:​Image-Based Search and Recommendation System
Problem Statement
Approach and Methodology
 Implementation
The Dataset
Installing and Importing Libraries
Importing and Understanding the Data
Feature Engineering
Calculating Similarity and Ranking
Visualizing the Recommendations
Taking Image Input from Users and Recommending Similar
Products
Summary
Chapter 6:​Pose Estimation
Top-Down Approach
Bottom-Up Approach
OpenPose
Branch-1
Branch-2
HRNet (High-Resolution Net)
Higher HRNet
PoseNet
How Does PoseNet Work?​
Pros and Cons of PoseNet
Applications of Pose Estimation
Test Cases Performed Retail Store Videos
Implementation
Step 1:​Identify the List of Human Keypoints to Track
Step 2:​Identify the Possible Connections Between the
Keypoints
 Step 3:​Load the Pretrained Model from the PyTorch Library
Step 4:​Input Image Preprocessing and Modeling
Step 5:​Build Custom Functions to Plot the Output
Step 6:​Plot the Output on the Input Image
Summary
Chapter 7:​Image Anomaly Detection
Anomaly Detection
Approach 1:​Using a Pretrained Classification Model
Step 1:​Import the Required Libraries
Step 2:​Create the Seed and Deterministic Functions
Step 3:​Set the Hyperparameter
Step 4:​Import the Dataset
Step 5:​Image Preprocessing Stage
Step 6:​Load the Pretrained Model
Step 7:​Freeze the Model
Step 8:​Train the Model
Step 9:​Evaluate the Model
Approach 2:​Using Autoencoder
Step 1:​Prepare the Dataset Object
Step 2:​Build the Autoencoder Network
Step 3:​Train the Autoencoder Network
Step 4:​Calculate the Reconstruction Loss Based on the
Original Data
Step 5:​Select the Most Anomalous Digit Based on the Error
Metric Score
Output
Summary
Chapter 8:​Image Super-Resolution
Up-Scaling Using the Nearest Neighbor Concept
Understanding Bilinear Up-Scaling
Variational Autoencoders
Generative Adversarial Networks
The Model Code
Model Development
Imports
Running the Application
Summary
Chapter 9:​Video Analytics
Problem Statement
Approach
Implementation
Data
Uploading the Required Videos to Google Colab
Convert the Video to a Series of Images
Image Extraction
Data Preparation
Identify the Hotspots in a Retail Store
Importing Images
Getting Crowd Counts
Security and Surveillance
Identify the Demographics (Age and Gender)
Summary
Chapter 10:​Explainable AI for Computer Vision
 Grad-CAM
Grad-CAM++
NBDT
Step 1
Step 2
Steps 3 and 4
Grad-CAM and Grad-CAM++ Implementation
Grad-CAM and Grad-CAM++ Implementation on a Single
Image
NBDT Implementation on a Single Image
Summary
Index
About the Authors
Akshay R Kulkarni
is an AI and machine learning (ML)
evangelist and a thought leader. He has
consulted with several Fortune 500 and
global enterprises to drive AI and data
science-led strategic transformations. He
is a Google developer, author, and regular
speaker at major AI and data science
conferences (including Strata, O’Reilly AI
Conf, and GIDS). He has been a visiting
faculty member for some of the top
graduate institutes in India. In 2019, he
was featured as one of the top 40 under
40 data scientists in India. In his spare
time, he enjoys reading, writing, coding,
and helping aspiring data scientists. He
lives in Bangalore with his family.

Adarsha Shivananda
is a data science and ML Ops leader. He is currently working on creating
world-class ML Ops capabilities to ensure continuous value delivery
from AI. He aims to build a pool of exceptional data scientists within
and outside of the organization to solve problems through training
programs, and he strives to stay ahead of the curve. He has worked
extensively in the pharmaceutical, healthcare, CPG, retail, and
marketing domains. He lives in Bangalore and loves to read and teach
data science.
Nitin Ranjan Sharma
is a manager at Novartis. He leads a team
that develops products using multi-
modal techniques. As a consultant, he
has developed solutions for Fortune 500
companies and has been involved in
solving complex business problems
using machine learning and deep
learning frameworks. His major focus
area and core expertise is computer
vision, including solving challenging
business problems dealing with images
and video data. Before Novartis, he was
part of the data science team at Publicis
Sapient, EY, and TekSystems Global
Services. He is a regular speaker at data
science community meetups and an
open-source contributor. He also enjoys training and mentoring data
science enthusiasts.
About the Technical Reviewer
Jalem Raj Rohit
is a senior data scientist at Episource,
where he leads all things computer
vision. He co-founded ML communities
like Pydata Delhi and Pydata Mumbai
and organizes and speaks at meetups
and conferences.
He has authored two books and a
video lesson on the Julia language and
serverless engineering. His areas of
interest are computer vision, ML Ops,
and distributed systems.
© The Author(s), under exclusive license to APress Media, LLC, part of Springer
Nature 2022
A. Kulkarni et al., Computer Vision Projects with PyTorch
https://doi.org/10.1007/978-1-4842-8273-1_1

1. The Building Blocks of Computer

Vision
Akshay Kulkarni1 , Adarsha Shivananda2 and Nitin Ranjan Sharma3
(1) Bangalore, Karnataka, India
(2) Hosanagara tq, Shimoga dt, Karnataka, India
(3) Bangalore, India

Humans have been part of a natural evolutionary pattern for centuries.

According to the Flynn Effect, an average person born in recent times
has a higher IQ than the average person born in the previous century.
Human intelligence allows us to learn, decide, and make new decisions
based on our learnings. We use IQ scores to quantify human
intelligence, but what about machines? Machines are also part of this
evolutionary journey. How have we moved our focus to machines and
made them intelligent, as we know them today? Let’s take a quick look
at this history.
A breakthrough came in the 1940s when programmable digital
computers became available, followed by the concept of the Turing test,
which could measure the intelligence of machines. The concept of the
perceptron goes back to 1958, when it was introduced as a powerful
logical unit that could learn and predict. The perceptron is equivalent to
a biological neuron that helps humans function. The 1970s saw fast
growth in the field of artificial intelligence, and it has increased
exponentially since that time.
Artificial intelligence is the intelligence showcased by a machine,
more often when it is trained on historical events. Humans have been
trained and conditioned their whole lives. We know, for example, that
going too near a fire will cause us to be burned, which is painful and
bad for our skin. Similarly, a system can be trained to make distinctions
between fire and water, based on the features or on historical evidence.
Human intelligence is being replicated by machines, which gives rise to
what we know as artificial intelligence.
Artificial intelligence encompasses machine learning and deep
learning. Machine learning can be thought of as mathematical models
that help algorithms learn from data/historical events and formulate
decision-making processes. The machine learns the pattern of the data
and enables the algorithms to create a self-sustaining system. Its
performance can be limiting, such as in the case of huge complex data,
which is where deep learning comes into the picture. Deep learning is
another subset of artificial intelligence. It uses the concept of the
perception, expands it to neural networks, and helps the algorithms
learn from various complex data. Even though we have many modeling
techniques at our disposal, it’s best to find good and explainable results
from the simplest of techniques, as stated by Occam’s razor (the
simplest answer is often the best).
Now that we have explored a bit of the history, let’s browse through
the applicable fields. There are two fields—Natural Language
Processing (NLP) and Computer Vision (CV)—that use an immense
amount of deep learning techniques to help solve problems. NLP caters
to the problem sets defined by our language, essentially one of the most
important modes of communication. CV, on the other hand, addresses
vision-related problems. The world is full of data that humans can
decipher with their senses. This includes vision from the eyes, smell
from the nose, audio waves from the ear, taste from the tongue, and
sensations from the skin. Using this sensory input, the connected
neurons in our brains parse and process the information to make
decisions on how to react. Computer vision is one field that addresses
the visual side of the machine-learning problem.
This book takes you through the fundamentals and gives you a
working knowledge of computer vision.

What Is Computer Vision

Computer vision deals with specific problem sets that rely on images
and videos. It tries to decipher the information in the images/videos in
order to make meaningful decisions. Just like humans parse an image or
a series of images placed sequentially and make decisions about them,
CV helps machines interpret and understand visual data. This includes
object detection, image classification, image restoration, scene-to-text
generation, super-resolution, video analysis, and image tracking. Each
of these problems is important in its own way. Studying vision-related
problems has gained a lot of attraction after the power of parallel
computing came into play.

Applications
The applications of computer vision vary with respect to the industry
being discussed. The following sections look at a few of these tasks.

Classification
The simplest form of an image-related problem involving a decision-
making process is a supervised technique, called classification.
Classification simply involves assigning classes to different images. The
process can be as simple as an image having one class or it can more
complex, when there are multiple classes within the same image. See
Figures 1-1 and 1-2.

Figure 1-1 The class in this case is a cat

Figure 1-2 The class in this case is a dog
We can separate the content of such images based on whether they
have an image of a cat or a dog. This is an example of how our eyes
perceive differences. The background of the object we are trying to
classify does not matter, so we need to make sure that it doesn’t matter
in the algorithms as well. For example, if we included a logo of some car
company in front of all the dog images, the image classifier network
might learn to classify dogs based on that logo and use it as a shortcut.
We later describe in detail how to incorporate this information into the
model. Classification can be used to identify objects in a production line
of a manufacturing unit.

Object Detection and Localization

An interesting problem that is often encountered is the need to locate a
particular image inside another image and even detect what that might
be. Let’s say there is a crowd of people and some are wearing masks
and some are not. We can use a vision algorithm to learn the features of
masks, then use that information to locate a mask relative to the image
and detect the masks. See Figures 1-3 and 1-4.
Figure 1-3 Class: No mask detected
Figure 1-4 Class: A mask is detected in the image
This analysis can be helpful in detecting license plates of moving
vehicles from traffic cameras. Sometimes, due to the resolution of the
cameras and the moving traffic, the picture quality is not that great.
Super-resolution is one technique that is sometimes used to enhance an
image’s quality and help identify the numbers on the plate.
Image Segmentation
This process is used to determine edges, curves, and gradients of
similar objects placed together in order to separate different objects in
an image. A classic unsupervised technique can be used here without
the worry of finding good-quality, labeled data. The processed data can
further be used as an input to an object detector. See Figure 1-5.
Figure 1-5 Separating terrains in topological maps

Anomaly Detection
Another classic, unsupervised way of determining changes is to
compare an image to the usual, expected patterns from some training
data. Anomaly detection can be used, for example, to determine
imperfections in steel pipes when compared to training data. If the
machine finds something odd, it will detect an anomaly and inform the
line engineers to take care of it. See Figures 1-6a and 1-6b.

Figure 1-6a Perfect examples of steel pipes

Figure 1-6b Anomalies showing up in the pipes

Video Analysis
There are a lot of use cases for video or sequences of images. The task
of object detection on running images can help with CCTV footage. It
can also be used to detect abnormalities within the frames per video
section.
We will be going through all of these applications in detail in the
upcoming chapters. Before that, let’s go through a few of the intrinsic
concepts that lay the foundation for further understanding computer
vision.

Channels

Figure 1-7 Playing musicians

One of the most basic and quintessential ideas around computer vision
is the channel. Think of music being played with multiple instruments;
we hear a combination of all the instruments playing together, which
essentially constitutes the music in stereo (see Figure 1-7). If we break
the music into single components, we can break the sound wave into
individual sounds coming from the electric guitar, the acoustic guitar,
the piano, and the vocals. After breaking the music into its components,
we can modulate each component to get the desired music. There can
be an infinite number of combinations if we learn all the musical
modulations.

Figure 1-8a Pixel values corresponding to a white picture

Figure 1-8b Sample white page

Figure 1-8c Representational pixel values corresponding to a sketch

Figure 1-8d Sample sketch represented in one channel
We can extrapolate these concepts to images, which we can break
into components of colors. Pixels are the smallest containers of colors.
If we zoom in on any digital image, we see small boxes (pixels), which
make up the image. The general range of any pixel in terms of the
intensity of a channel is 0-255, which is also the range defined by eight
bits. Consider Figure 1-8b. We have a white page. If we convert that
page to an array, it will give us a matrix of all 255 pixels, as shown in
Figure 1-8a. On the other hand, Figure 1-8d, when converted to a
matrix, will also have only one channel, and the intensity will be
defined by the numbers in the range of 0-255, as depicted in Figure 1-
8c. Closer to 0 gives us black and closer to 255 gives us white.
Let’s consider a color image. We can break any full-color image into
a combination of three primary components (channels)—red, green,
and blue. We can break down any color image into some definite
combination of red, green, and blue. Thus, RGB (red, green, and blue)
becomes the channels of the colored image.

Figure 1-9 Sample image to blue (left)=0, green (middle)=1, and red (right)=2
 The image in Figure 1-9 can be split into RGB, with the first channel
being blue, then green, followed by red. Each pixel in the image can be a
certain combination of RGB.
We are not restricted to using RGB as the color channels. We also
have HSV (Hue, Saturation, and Value), LAB format, and CMYK (Cyan,
Magenta, Yellow, and Black), which are a few representations of
channels of an image. Color is a feature and its container is a channel,
so every image is made of edges and gradients. We can create any
image in the world with just edges and gradients. If you zoom in on a
small circle, it should look like a combination of multiple edges and
straight lines.
To summarize, channels can be called in as the container for a
feature. The features can be the smallest individual characteristics of
the image. Color channels are a specific example of channels. Since
edges can be features, a channel that only caters to edges can be a
channel too. Food for thought—if you were to create a model that
would identify a cat or a dog, does the color of the animal affect the
model behavior as much as edges and gradients can?

Convolutional Neural Networks

You now know that images have features, and those features need to be
extracted for a better understanding of the data. If we consider a matrix
of pixels, the pixels are related in all four directions. How do you do the
extraction efficiently? Will the traditional methods of machine learning
or deep learning help? Let’s go through some problems:
1. Images can have huge dimensions. For example, a 2MP image, if it is
allowed to capture a 1600x1200 image, will have 1.9 million pixels
per image.

2. If we are capturing the data via the images, the data is not always
centrally aligned. For example, a cat can be in one corner of one
image, and on the next image, it can be in the center. The model
should be able to capture the spatial changes in the information.

3. A cat in an image can be rotated along the vertical dimension or the

horizontal dimension and still it remains a cat. Thus, we need a
robust solution to capture such differences.
 We need major upgrades from our regular tabular data approach. If
we can break down a problem into smaller manageable pieces,
anything can be solved. Herein, we use convolutional neural networks.
We try to break the image into several feature maps via kernels and use
those in sequence to build a model that can then be used for any
downstream or pretext task.
Kernels are feature extractors. Features can be edges, gradients,
patterns, or any of the small features discussed earlier in this chapter. A
square matrix is generally used to operate a convolution task in the
image on the first step and on feature maps from the second step. The
convolution tasks performed by the kernels can be thought of as the
simplest tasks in a dot product. See Figures 1-10a and 1-10b.

Figure 1-10a 3x3 matrix of feature maps

Figure 1-10b 3x3 kernel

Figure 1-10a is the image or feature map and Figure 1-10b is the
kernel. The kernel is the feature extractor, so it will do a dot product on
the feature map, resulting in the value 10. This is the first step of our
convolution. The images or feature maps are going to be large and so
the kernel might not operate on just a 3x3 matrix, but it will take a
stride forward to calculate the next convolution operation. Let’s look at
an extrapolated example of this idea.
Figure 1-11 Feature map, kernel, and resulting output
As shown in Figure 1-11, a 5x5 feature map was convolved by a 3x3
kernel and that resulted in a 3x3 feature map. That map will again be
convolved or converted to some features for a downstream task.
The convolution process also contains a concept of stride, which is a
hyperparameter telling the kernel how to move around on the feature
maps. Given our convolutional neural network, we have a stride of 1. A
stride greater than 1 can cause a checkerboard issue in the feature map,
with a few pixels getting more attention than the others. We might want
or not want this effect, based on our business requirements. A higher
stride value can also be used to lower the feature map dimensions.
There is an inherent problem with this convolution. The dimension
keeps on shrinking when the convolutions happen. This can be
desirable in some sense or some particular use cases but in a few use
cases, we might want to retain the original dimensions. We can use the
concept of padding on the image or the subsequent feature maps to
avoid the issue of dimensionality reduction. Padding is another
hyperparameter. We add layers around the images or feature maps.
Figure 1-12b shows how the padding essentially increases the space
as well as allows the edge pixel values to be better noticed by the
kernels. The convolution process will go through the edge values more
than once, thus the information is being carried forward to the next
feature maps with more effect. In the case of edges having pixel values,
the convolution from kernels will take that value only once, whatever
the stride values are on that occasion.
Figure 1-12a An edge detector

Figure 1-12b A zero-padded feature map

It is quite interesting how simple padding can change the
identification of edges by a kernel. If we assume there is an important
edge near the corner of the feature map and don’t pad it, the edges
won’t be detected. This is because, to detect a line or an edge, the kernel
or the feature extractor needs to have a similar pattern. In the case of
the kernel in Figure 1-12a, which is an edge detector, it needs to find a
proper gradient to detect an actual edge. It is because of the gradient
from 0 to 4, 0 to 5, and 0 to 5, now it detects the edge. If the padding
wasn’t there, this gradient wouldn’t be there and the kernel would have
missed out on an important part.

Receptive Field
When we were working on the concepts of convolutions, there was a
feature map and kernel stride. The kernel was extracting features in the
space such that the model could interpret the information to its ease.
Now we’ll consider with an example an 56x56 image with one channel.
(This is also written as 56x56x1.) If we try to convert the entire image
to features, the entire span of pixels needs to be read. Let’s look at a
graphic example to sort out this concept.
Figure 1-13 Diagram showing convolution operation
Figure 1-13 shows a normal 5x5 feature map being convolved by
another 3x3 kernel, resulting in a 3x3 kernel, assuming no padding and
the stride to be equal to 1. The information carried forward in the step
is being carried from the first block of the feature map to the next block
of the feature map. That means only the left-most highlighted corner
pixel information in Figure 1-13 is contained in the left-most pixel value
shown in the 3x3 feature map. This poses a problem because, if we have
only one layer and we have a 3x3 kernel, essentially one pixel will have
a receptive field of 3x3 in that layer. Interestingly, in the layer after that,
if we are using a 3x3 kernel again with stride 1 and zero paddings, the
receptive layer is again 3x3 for Figure 1-13, but the 3x3 has already
seen the whole image. Thus, it carries all the information from the
original image. This concept is called the local and global receptive
field.

Local Receptive Field

It is the layer-wise information strength that has been carried forward
by the convolutions by the kernel for that step. In the example shown in
Figure 1-13, we have a 3x3 kernel, or nine pixels. It will depend on the
immediate step and not the entirety.
Global Receptive Field
This is the cumulative information being carried to the last layer of the
model. Generally, we want to average out the information from the
image to just one value at the end for the sake of simplicity. If we have
to predict cats or dogs, it will be easy if we just get one value for the
same, rather than a matrix at the end. The last value (1x1xn) or pixel
should essentially see all the pixels of the image at the start to be able
to perfectly decipher the information.
The example in Figure 1-13 is being convolved by 3x3 and again by
3x3 to reach the final 1x1. Now the 1x1 would have seen the 3x3 figure
and that in turn would have seen the 5x5 figure. It’s passing the baton
and the kernel is helping. The global receptive field here is 5x5.

Pooling
An initial advantage of the convolutional neural network is that it works
with parallelism. A fully connected layer would have to worry about an
input vector of 25 dimensions even if the image size would be as small
as (5x5x1). Even though CNN solves this issue by working on spatial
fields, the high dimensionality still can cause the CNN architecture to
have a huge number of parameters. Pooling tries to solve this issue by
helping with a dimensionality reduction technique and filtering
information.
There is an inherent problem with just applying convolution layers
in the CNN. The spatial features are captured by the kernel, but a small
change in the input feature map will have a large impact on the output
feature map. To avoid this challenge, we can use pooling. Depending on
the downstream tasks we are doing, we can use max pooling, average
pooling, or global average pooling.
Pooling can be thought of as something similar to how the
convolution layer applies to the feature map. However, instead of
convolution, pooling will work on calculating an average or max of all
the values in the region. It can be thought of as a function. There are no
parameters that are learned in the pooling section. It is just plain and
simple dimensionality reduction in space. It is advisable to use a 2x2
pooling on a higher dimensional feature map, which needs to be greater
than 10x10 at least. The reason being the information concentration at
this level of a CNN will be too high and a rapid reduction of it by pooling
can result in a huge loss.
Max Pooling
Feature maps will contain the information from the image distributed
spatially. If we consider a downstream task that needs only the edges,
we can try to maximize the edges such that the subsequent feature
maps will have a piece of filtered information to focus on. When max
pooling has been selected, the prominent features are filtered and
passed on to the next layer. When the stride is looking into the stretch
of pixels, it selects the one with the highest value, thus the word max.

Figure 1-14 Example of max pooling

Figure 1-14 shows a feature map on the left; we are trying to get a
max pool with a stride of 2. The resultant feature map is reduced from
4x4 to 2x2, with just having the prominent feature passing over. In a
way, if there are some edges or gradients in an image or feature map,
that will take precedence over anything else.
In the case of tasks like classification, we generally follow the
process of max pooling, because we need the important edges and
gradients to follow and no other irrelevant features messing up the
architecture. It is interesting to ponder the fact that color doesn’t play a
role in most of the classification tasks. For example, a cat can be any
color and the model has to understand it is a cat without considering
the color.
Average Pooling
We have already established the basic concept of pooling. It gives us a
filtration process without increasing any number of learnable
parameters. For an average pooling of 2x2 and a stride of 2x2, the
strides taking care of a 2x2 block at a time will average the entire
section and calculate a mean on top of it. The mean is passed onto the
next section of the feature map.
For classification tasks, average pooling is generally not suggested.
It can however be used when the images are darker and you want to
extract black-to-white transitions.

Figure 1-15 Example of average pooling

Figure 1-15 shows a feature map being pooled by a 2x2 block with a
stride of 2. The average of the pooled area is reflected with each 2x2
block from the feature map shown on the left, mapped to one pixel on
the right side.
Global Average Pooling
Global average pooling is sometimes used toward the end of the CNN
architecture to summarize the feature map to one value. Assume, if you
were left with a 5x5 feature map with some depth (channels in the z-
direction, assuming x and y are height and width of the feature maps),
you could flatten those values out and use a fully connected network to
map those like features to get a model. This is a viable option, but
taking 5x5, we are forwarding 25 features to a fully connected network
and this will take up huge parameters. Instead, you can try to use a
global average pooling layer after this step and make it to 1x1,
essentially managing all the important features and having a
combination. The model parameters we saved by not using a fully
connected network can be used to add convolutional layers. This can
result in better accuracies.
Figure 1-16a Sample feature map

Figure 1-16b Result after GAP

Figure 1-16a shows a feature map with 4x4 dimensions. After
pooling, we are left with one single value, as shown in Figure 1-16b. In
all cases, we will have a depth, which will have values based on
corresponding pooling and feature maps.

Calculation: Feature Map and Receptive Fields

The calculation serves as an important aspect of this coursework. We
will be using it to formulate our models and various experimentations.
The dimensions of the output feature maps depend on multiple factors,
such as stride, kernel size, pooling, padding, input, and output. Let’s
dive into the details.
Kernel
The kernel is the feature extractor from the feature map or the image. It
is initialized in the first forward propagation but the weights are
expected to learn and change from the backpropagation to enable it to
be a better feature extractor. Once the training sets in, the weights move
toward the higher end. This can mean the features it’s extracting are
important to the cost function and thus the weights. Let’s denote the
dimension of the kernel with K.
Stride
The step by which the kernel moves around the feature map can be
called a stride. Let’s denote its value by S.
Pooling
The block in a conventional convolutional neural network architecture
tries not to convolve, but aims to capture some kind of information
from the spatial feature maps and rapidly reduce the dimension of the
feature map for the next step. Let’s denote that by mp.
Padding
Padding allows us to keep the dimensions constant after the
convolution process has taken place. Let’s denote that with P.
Input and Output
Let’s denote the input and output as is, where input refers to the
feature map at the first step and output refers to the feature map
generated.
The first step of getting an architecture started for an image will be
to access the size of the image and determine how deep we want to
build the network such that the receptive field of the last layer or the
output will be equal to the size of the image. In other words, it should
have seen what is there in the image to answer anything about the
image. If we have an image of size 56x56x3 the kernel to be used can be
3x3x3x16. This means the 3x3 kernel has three different sets of
initializations and matches up with the channels in the input or
preceding layer. This is because it has to know the possible way to mix
the features that need to be extracted and used in the network. The
formula becomes:
HxWxC > KxKxCxCnext
Here H, W, C are the height, width, and channel of the input
image/feature map.
K is the kernel size and Cnext is the number of channels in the next
step or the batch size. When using CUDA scores to train the CNN, we
need to push for the kernel size to be equal to some number defined by
2^n. For example, if we are using 17 as the number of kernels, it will
still end up using 32 instead of the closest 2^n number, i.e., 16.
Given all the values and concepts, let’s calculate the feature map
outputs:
Given, Input = 12x12,
P = 0, K = 3x3, S = 1

Output = 10 or 10x10
Calculation of Receptive Field
The receptive calculation at the nth layer will be given by a consolidated
formula, as follows

Let’s say we are calculating the receptive field at the second layer So
K1, K2 = 3.
The stride of the kernel is 1 for both occasions and the receptive
field comes to 5.

Understanding the CNN Architecture Type

Understanding Types of Architecture

AlexNet
ILSVRC is a competition that was dedicated to computer vision research
with its famous ImageNet dataset being used to evaluate models and
research in the field. One of the first winners was AlexNet, which had
great accuracies in 2010 and 2012. The paper released corresponding
to this network revealed usage of the convolutional network, with basic
building blocks.
Figure 1-17 AlexNet architecture
Figure 1-17. depicts the overview architecture used in AlexNet. The
image size taken was 224x224x3 and brought down by a 11x11x3x96
convolution with a stride of 4 to 55x55x96. The max pooling with 3x3
filters and a stride of 2 was used. The first two convolutional layers
used LRN and pooling. Following that, the next three layers only used
the convolutional and activation layer, followed by two blocks of the
fully connected network. The scores were then passed onto the softmax
to classify the 1,000 classes.
The most interesting invention in the model architecture is LRN
(Local Response Normalization). At the time, sigmoid and tanh were
generally used as activation functions, but AlexNet used ReLU. Sigmoid
and tanh suffer from saturation at extreme values and the data always
has to be centered and normalized to get any gradient pass on during
backpropagation. The local response normalization can be thought of
as a brightness normalizer and is followed by the ReLU activation.
AlexNet tries to use multiple GPUs to parallelize the training
process with increased accuracy. There are two parts to the network
that are parallel to each other crossing over at certain sections.
VGG
Figure 1-18 VGG architecture

Figure 1-18 shows conv 1, conv 2, conv 3, conv 4, and conv 5—a
convolution block with ReLU activation. Each block is followed by a max
pooling followed by FC 6, FC 7, and FC 8—a fully connected network.
Figure 1-19 Layer-wise depiction of a VGG stack
The architecture started a long-drawn application of going deep
with a basic development style. At the time this architecture was
proposed, batch normalization was not used so the network suffered
from internal covariate shift and gradient lost midway. The architecture
uses a combination of 3x3 and followed by 1x1. The reason that
stacking three 3x3 is better than one 7x7 is in terms of parameters; if
we have C channels, then 3x3 has 27C2, but in the case of 7x7 we will
end up having 49C2. However, the receptive field ends up with three
stacks of 3x3, which is equal to one 7x7, all with more adaptive
functions convolving over the feature maps. See Figure 1-19.
 The model architecture follows a convolution block-wise
architecture that uses a well-defined precedence, used in modern
architectures. We have five convolution blocks, and when the feature
maps have reduced size, this keeps the information loss minimal. The
number of channels increases at each block. Each convolution block is
followed by max pooling to reduce the feature maps’ dimensions. There
is also an addition of 1x1 convolution, which acts as a mixer of all the
features of the feature map from the preceding level. This convolution
acts as a z-axis dimensionality reduction technique. While continuous
usage of 3x3 or 5x5 will increase the number of channels for a feature
map, at some point, it needs to get reduced, with proper technique
keeping the information loss minimal. For all purposes, 1x1 is used as a
cross-channel pooling. It can also be used to increase the number of
channels but it is not done in practice. Since it is element-wise
multiplication on the feature maps it can easily summarize the content.
VGG was a big improvement over the conventional CNN architecture
and was the state-of-the-art algorithm published in 2015.
ResNet
ResNet is an advanced architecture for image downstream tasks and
made its a grand entry into the world of CV by winning the ILSVRC
competition in 2015. It is also the backbone for a few of the state-of-
the-art object detection algorithms like YOLO and faster RCNNs. The
paper was put forward by a team from Microsoft. They were able to
train deeper with the help of the residual learning framework.
Figure 1-20 Residual structures
Figure 1-20 shows the basic structure of the residual framework.
The output from the previous layer can be considered as X; it is being
passed to the residual function along with an identity function. So, if the
residual function is given by the function f(x), we can call the result
from both f(x) + x.
The issue that the structure was trying to solve was a degradation of
accuracies. The experiments revealed the deeper networks were having
their accuracies saturated after the addition of layers and finally falling
off. These drops were not from overfitting but due to the problem itself
being hard to optimize. In deep neural networks, a common linear
problem can be hard to train. For example, if want to build a model with
just a linear addition of two numbers and we want the resultant to be
the sum, a non-linear counterpart is often easy to optimize, like just
adding exponentials to the numbers and the resultants.
Figure 1-21 ResNet basic architecture

Figure 1-22 ResNet layer-wise convolution information

The identity layers that result from skipping the residual networks
provide a second set of available local receptive fields in addition to
what we are getting from the residual network. Often termed as the
skip connections or the highway networks, they provide a copy of the
image to the deep nets. The output from the skip connection and the
residual need to have the same dimensions. Thus there is a projection
that happens on the output to match the dimensions before being
added to the residual functions. Figure 1-21 depicts a ResNet-34
architecture, which can be referred from the table shown in Figure 1-
22. The architecture uses a stack of 3,4,6,3, a stack of 64,128,256, and
512 network blocks. There are two types of skip connections—one
with solid lines and the other with dotted lines. In most cases if the
input dimensions match the output of the residuals, the input is added.
This is shown by the solid lines. In the case of dotted lines, the
dimensions, if not matched, are controlled by padding in addition to the
1x1 with stride 2. Either way, no parameters are being added for the
model to learn.
Another important point to note is that this model uses a stochastic
gradient descent with momentum as an optimizer. SGD with
momentum is another proven method used in a lot of other state-of-
the-art models.
Inception Architectures
Inception has a few versions delivered since 2014. GoogLeNet (often
linked with InceptionV1) was the winner of ILSVRC. There are multiple
iterations; let’s look at inception architecture.
To increase effective local representation in the convolution block,
this solution proposed to factorize convolutions. Effectively it suggested
using 3x3 after 1x1, reducing the intensity of the activations and
reducing correlations. They used a way of factorizing heavier
convolution kernels to smaller counterparts trying to achieve the same
result. For example, a 5x5 requires 25C2 parameters but it can be
broken into two sets of 3x3 and still require 18C2 parameters. See
Figure 1-23. According to the research, 5x5 might have better
expressiveness in terms of getting hold of more pixel points in one
convolution. Thus, it has a better bird’s eye view compared to 3x3.
However, the architecture claims that while building a computer vision
model, we need to care for translational invariance and a smaller kernel
size will help us with that.

Figure 1-23 Bird's eye view of convolution

The factorization is further broken into asymmetric convolutions.

The 3x3 can still be broken into 3x1 and then 1x3 convolutions,
resulting in at least 9/6 savings. This can be used in a higher set of
convolutions as well and is shown generically in Figure 1-24.
Discovering Diverse Content Through
Random Scribd Documents
"Indeed, then, you are finely mistaken. I'll be nothing of the kind."
"You've not heard what I'm going to say, yet. Won't you sit down
and let us be comfortable?"
"You can sit if you fancy it. I prefer standing."
"Very well; we shall both be pleased. This is a very comfortable
chair."
Janet sank back in it, and raised her placid face to Bridget's. The two
girls were in all particulars contrasts. Biddy's curls were now a mop;
a wild, aggressive, almost disreputable looking mop. Her white dress
was draggled and crumpled, her cheeks were deeply flushed, her
eyes flashed ominous fire, her proud lips took many haughty and
defiant curves. Janet, in contradistinction to all this, was the soul of
neat commonplace. Her pale blue cambric frock fitted her neat figure
like a glove. She had white linen cuffs at her wrists; her little hands
were exquisitely clean; her fair face looked the essence of peace.
Her neat, smooth head of light hair shone like satin.
"I am anxious about you," said Janet. "I can see quite plainly that
you are going all wrong."
Bridget gave a sort of snort.
Janet held up her small hand imploringly.
"Do listen," she said. "How can I explain myself if you interrupt me
each moment?"
"But you never liked me, Janey. You have shown that all too plainly.
I cannot imagine what you are prying into my affairs for. Now if
Dolly came——"
"Dolly has not come, and I have. Now, will you listen. I will frankly
say that I did not care about you when you first came to the school.
When I saw you so—so defiant, Bridget, so proud, so free, so
absolutely fearless; when I saw you with all these characteristics,
taking people by storm, for you know you did take the little girls of
the school quite by storm, I felt a sense of strong irritation against
you. I never met a girl like you before; you puzzled me; you did not
please me. Now, I am going to be quite frank; I do not really like
you much better now, but as I see that you fully intend to be on my
side, it is impossible for me any longer not to take your part."
"I fully intend to be on your side?" repeated Bridget. "Indeed, then, I
don't, and I may as well say so frankly at once."
"Yes, Bridget, you do; you can't help yourself, for you and I will in
future have good cause to hate the same girl."
"What girl?'
"Evelyn Percival; the one you have just spoken of as the queen of
the school."
"The darling!" exclaimed Bridget, "and why in the name of goodness
am I to hate her?"
"Well, you must be a poor-spirited thing if you don't. May I ask if you
would have got into your present scrape but for her? Have you not
before this disobeyed Mrs. Freeman? Up to last night she took pity
on you; she said to herself: 'Bridget knows nothing of the rules of
the school; Bridget has never been accustomed to obey any rules, I
will be merciful to her, I will be lenient, I will never forget that Biddy
has been queen in her Irish home.'"
"Oh, don't talk to me about my home," said Bridget, her lips
quivering, her eyes filling with tears.
"Yes; but is it not true, Bridget? Has not Mrs. Freeman been very
lenient to you in the past?"
"I suppose she has. I never thought much about it. I scraped along
somehow; I was happy enough."
"Well, was she lenient to you to-day?"
"Need you ask, Janet? I'm a prisoner; a close prisoner in this
abominable room. Such treatment will soon kill me. I can't eat; I
shall soon die of misery."
"It is hard on you, Bridget; you are exactly like a wild bird of the
woods put into a cage."
"Yes, that's it; and the captive bird will break its heart."
"Poor Bridget! I didn't like you in your free days, but I'm willing to
own that I pity you now."
"Thank you, thank you; but I hate pity. Whoever would think of
offering pity to Bridget O'Hara at home?"
"But Bridget O'Hara is no longer at home; she is a captive in a
strange land. Don't cry, Biddy. Let us leave sentimentalities now, and
come to facts. Whom do you think you owe this severe treatment
to?"
"I am sure I can't tell you."
"I can tell you, however. You owe it entirely—to Evelyn Percival."
"Now what do you mean? that nice girl whom I nearly killed?"
"You didn't nearly kill her; that's all stuff! Bridget, you don't know
Evelyn Percival, but I do. Had any other girl been in the carriage
when you and the children startled the horses, you would have been
forgiven. Mrs. Freeman would still have remembered that you were
unaccustomed to rules, and she would have tried to break you in
gently and considerately; but as Evelyn happened to be the person
whose delicate nerves sustained a shock, Mrs. Freeman was
incapable of showing any mercy. Evelyn Percival poses in the school
as a sort of saint. Nearly everyone bows down to her; Mrs. Freeman,
head mistress though she is, is so influenced by her that you are
sure to have a bad time in future."
"I shan't stand it; it isn't likely."
"You will be forced to stand it. If Evelyn gives the smallest
suggestion about you, it will be certain to be followed out. I pity you,
Bridget, but you are certainly likely to have a lively time."
"You don't mean to tell me," answered Bridget, "that I have to thank
Miss Percival for this punishment; that it is at her instigation I am
here?"
"You are certainly here at no one else's instigation."
"Did she tell Mrs. Freeman to make a close prisoner of me, and to
starve me?"
"It is your own fault if you are starved, Bridget; don't exaggerate,
my dear; you do no good by that. As to your being made a prisoner,
you certainly owe it to Evelyn. She can say things, even though she
does not put them into words."
"Oh, I understand," said Bridget. She turned again to look out of the
window, and her impatient fingers once more played a tattoo on the
glass.
"Evelyn is most popular," continued Janet, "for the simple reason
that people don't read her through and through. I can see beneath
that sweet, saintly calm, and I honestly say that I cannot bear her.
Now, Bridget, if you will come on my side, if you will join me in
opposing the pernicious influence that girl exercises, I can help you
out of this scrape without allowing you to humiliate yourself, and I
can at the same time put you up to having the nicest little revenge
in the world on this delightful Miss Percival."
"But Dorothy believes in her, and Dorothy is so sweet and kind,"
exclaimed Bridget, in perplexity.
"Poor, dear Dolly," exclaimed Janet, "anyone can take her in; but
you, my dear, although you are not very learned, are clever.
However, this is your own concern. If you like to stay in this hot
room until Mrs. Freeman breaks in your proud spirit, and if you like
to submit to the many indignities which I can plainly see are before
you, that, of course, is your affair. I thought it only kind to warn you,
but perhaps I have interfered unwarrantably. If so, forgive me."
Janet rose as she spoke, and took a step or two toward the door.
"No, don't go," exclaimed Biddy. "You puzzle me very much; there's
no one in the world who hates mean ways more than I do, and if
Evelyn is that sort——"
"She is that sort, Bridget."
"Well, well!" Bridget turned again to the window.
"What am I to do, Janet?" she said, after a pause. Her tone was
quite humble; there was a crushed expression in her face.
"Poor old thing!" said Janet, in her light, silvery voice. She went up
to Bridget, and gave her a careless kiss on her cheek. She could
afford to do this, for she knew the victory was hers.
"In the future I will be your friend," she said; "you may rely upon
me. We are going to choose fresh chums in a week's time. Suppose
we choose one another. I know we are not a bit alike, but that's just
the very thing; opposites should keep together. However, there's
time enough to settle that presently."
"Yes, quite time enough," said Bridget. "I thought that I'd take Dolly
for my chum."
"You can't get her, my dear; she's bespoken to Evelyn long ago."
"That horrid Evelyn!" Bridget stamped her foot impatiently.
"Ah, I see, Biddy, that you and I will get on capitally. I could kiss you
again, but kissing isn't my way. Now then to business. The first thing
is to get you out of this room."
"How is that to be effected? Mrs. Freeman says that I am to stay
here until I promise to obey the rules of the school. I can't obey
them, so I suppose I'm to stay here until I die."
"And why can't you obey them, Bridget?"
"Why can't I obey the rules of the school? We are not likely to be
chums if you talk to me in that fashion, Janet."
"Now, my dear, I must just reason with you a little. You say you can't
obey the rules of the school; you say so because you fail to
understand them. If you put yourself under my guidance, and I am
quite willing to take charge of you, I will show you that you can obey
them sufficiently to keep yourself out of all serious scrapes, and yet
at the same time you will enjoy as much liberty as any girl need
desire. Do you think I am unhappy on account of the rules of the
school?"
"No; but you haven't got a wild heart like me."
"Poor Biddy, I'll take care of your wild heart. It was ill-natured of me
not to see after you before, but in the future, my dear, you are quite
safe. I am going to fetch Mrs. Freeman now."
"What in the world for?"
"To tell her that you will obey the rules, that you will cease to be an
unruly member of the community, that you are going to be my
chum."
"O Janet, but it's dreadful to promise and not to perform. I have
been awfully naughty, I know, over and over and over again, but I
have never stooped to breaking a promise."
"You shall not break this promise, for I won't let you, but I can show
you a way to keep the fetters from galling. Now I am going to fetch
Mrs. Freeman. It's worth your while to submit at once, Biddy, for I
intend to take you for a row."
"A row on the water!" Bridget's eyes sparkled; she threw back her
shoulders with a gesture of relief.
"Yes," repeated Janet, "a row on the water. The school boat is at our
disposal this evening. Mademoiselle is coming to take charge of us,
but, as she is really nobody, we shall practically be as free as air.
Stay where you are, Biddy, until I fetch Mrs. Freeman."
 CHAPTER IX.
TAKING SIDES.

When Dorothy entered Evelyn's bedroom she found her friend up

and dressed.
"I'm quite well, really, Dolly," said Evelyn, with a smile. "I stayed in
bed until I could endure it no longer. I can't tell you how vexed I am
that I fainted last night, and gave Mrs. Freeman a fright. There was
nothing really to make anyone else faint, for that brave girl saved
me from being hurt in the most wonderful manner. By the way, how
is she? I should like to see her and to thank her."
"Poor Eva," said Dorothy, coming up and kissing her friend, "you are
just the most forgiving creature in existence. Anyone else would be
awfully angry with Bridget. Her conduct very nearly cost you your
life!"
"There is a wide difference between 'very nearly' and 'quite,'" said
Evelyn, with a smile. "I escaped with a 'very nearly,' and feel as well
as ever now, and rather ashamed of myself. There never was a girl
who meant less harm than this Bridget. I can see her now running
down the road, her face all smiles, her eyes dancing, her white teeth
showing; I can see the little ones surrounding her. They waved
boughs of trees, and they shouted and sang as they came. For one
moment I said to myself, 'O Jubilate! here is a welcome worth
having!' but then Caspar took fright, the carriage swayed horribly,
the cushions jumped up as if they were going to strike me, and I
remembered nothing more until I awoke with my head on this girl's
lap, and Mrs. Freeman bending over me. I should like to see the girl,
to thank her. Where is she, Dolly? I am attracted by her face; it is a
very lovely one!"
"Well, sit down, now, by the window, and let us talk," answered
Dorothy. "I shall be jealous if you give all your thoughts to Bridget
O'Hara. I know she's a pretty girl, and I like her very much for some
things. But, oh dear, she is a care! I don't believe that any school
had ever before such a madcap in it. But don't let us waste all our
time talking about her. You can't help hearing her name spoken
morning, noon, and night, when you come into the school."
Evelyn sank down in a low easy-chair by the open window. She wore
a white cambric dress, and a pale blue belt round her slender waist.
Her gentle eyes, also faint blue in their coloring, looked out over the
summer scene. She was not beautiful, but there was a charm about
her, a sense of repose, which made it delightful to be with her. The
singular unselfishness of her nature was apparent in everything she
did, said, and thought.
"I'm delighted to be back, Dolly," she said. "This illness of mine has
been such a bother, and it's delicious to be well and able to go in for
things again. Now, if I may not speak of Bridget, tell me about the
other girls in the school. Tell me, also, what is the great object of
interest at present?"
"Oh, the Fancy Fair!" Dorothy colored as she spoke. "You need not
bother your head about it, Evelyn," she continued quickly. "Janet is
at the head of it; it was she who thought of the fair, and she's the
moving spring. You know what that means, don't you, darling?"
"I'm afraid I do," replied Evelyn. "Does Janet May dislike me as
much as ever?"
"She certainly does; but don't fret about her; she's not worth it. Eva,
you will most likely be asked to come on the committee, and to take
a stall at the Fancy Fair. If you get the invitation, will you accept it?"
"Of course I shall. Need you ask? Alack and alas! I have no chance
of winning any prizes, so the fair will be a great diversion. I suppose
it's a charity concern; who is it for?"
"A little orphan boy in the neighborhood. Oh, you'll learn all about
him presently. We are working as hard as possible for the fair. If you
come on the committee, Evelyn, you must let me help you with your
stall."
"If I come on the committee," repeated Evelyn. "I suppose I am
quite certain to be asked to join? Dolly, you look at me in rather a
queer way!"
"Do I? Don't notice my looks. There is something worrying me, but
nothing bad may come of it. It is so nice to talk to you again. Now I
have something to say about that poor Biddy. At the present
moment she is in disgrace."
"In disgrace? What about?"
"I'm afraid it's about you."
"Oh, but I must speak to Mrs. Freeman. She really meant nothing
wrong, dear child."
"She broke the rules in leaving the grounds without leave. I think it
is for her disobedience that Mrs. Freeman is punishing her. She has
shut her up in Miss Patience's room, and poor Biddy won't eat, and
is in a dreadful state of mind. Marshall spoke to me about her after
dinner, and asked me to go to her; but we had a committee meeting
just then, and afterward I could not find Mrs. Freeman."
"Have you left the poor girl by herself all this time, Dolly?"
"I must own that I have. I will go and have a talk with her as soon
as ever I leave you; not that I can do much good, she's such a
queer kind of mixture of obstinacy and passion."
"But it does seem dreadful to leave her by herself all this time; just
as if no one had a scrap of sympathy for her. Let us both go to her
at once, Dolly. I want to thank her for being so brave."
"But Mrs. Freeman; we ought to ask her leave."
"Mrs. Freeman will be in her own sitting room at this time. Come
along, Dolly, we have just a few minutes to spare before the gong
sounds for tea."
Dorothy made no further objections, and she and Eva went
downstairs side by side.
They knocked at Mrs. Freeman's sitting-room door. She was not in,
but Miss Delicia was tidying books and papers on her davenport.
"Is that you, Eva!" she exclaimed in delight. "Why, you look as well
and jolly as possible. How nice to have you back again!"
The little lady ran up to Evelyn, and kissed her affectionately. "Now,
my darling, you are not going to tire yourself," she said. "Come and
sit here by the open window."
"I have been sitting still and lying down all day," replied Evelyn, with
a faint little grimace; "I am not really tired at all. Dolly and I came,
Miss Delicia, to ask Mrs. Freeman to give us leave to go and see that
poor girl, Bridget O'Hara. It seems she has got into a scrape on my
account."
"And rightly, my dear; and very rightly. For my part, I don't approve
of punishments; I am all the other way; but such conduct as
Bridget's does deserve a sharp reprimand. Suppose you had been
seriously hurt, Evelyn?"
"But I was not hurt at all. I wish I could go and see Miss O'Hara
now; I want to thank her for having saved my life. If she did give me
a fright, Miss Delicia, she also kept me from the consequences of her
own act. I wish I could thank her."
"Well, dear, do go to her; I'll give you permission, and set things
right with Mrs. Freeman. If you and Dolly can bring that wild child to
hear reason we shall all be only too delighted. Run away, my dears,
both of you, and do your best."
The girls left the room, and ran down the stone passage which led
to Miss Patience's little sitting room at the other side of the big
house.
They were surprised, however, on reaching it, to find the door flung
wide open and the room empty.
Dorothy gave an exclamation of astonishment.
"Bridget must have given in," she said; "Mrs. Freeman must have
come to her, and she must have yielded. Oh, what a relief! How glad
I am! Come, Evelyn, let us go on the terrace, and walk up and down
until tea is ready."
The broad terrace which ran in front of the house was completely
sheltered from the sun at this hour. There was a pleasant breeze,
and the girls, as they paced arm in arm up and down the broad
path, looked happy and picturesque.
Two girls who were coming up the grassy slope at this moment
stopped at sight of them; one uttered a slight exclamation of dismay,
the other made an eager bound forward.
"There's Dolly!" exclaimed Bridget; "do let me run to her, Janet."
"Miss Percival is with her," exclaimed Janet. "Do you really want to
speak to Miss Percival, Bridget, after all you have suffered on her
account?"
"But she looks very nice."
"What a poor, weak kind of creature you are to be influenced by
looks; besides, she is in reality very plain. Even her warmest
admirers have never yet bestowed on her the palm of beauty."
"Oh, I like her face; it looks so good."
Janet paused in her walk to give her young companion a glance of
steady contempt.
"Can I possibly go on with this scheme of mine?" she muttered to
herself. "Bridget O'Hara is altogether too dreadful." Had Janet
yielded to her impulses at that moment she would have told Bridget
to join her beloved Dorothy and Evelyn Percival, and have declared
her intention of washing her hands of her on the spot. Had Janet
acted so, this story need never have been written. But that strong
ambition, that thirst for praise, which was her most marked
characteristic came to her aid. Bridget was the only means within
her power to achieve a most desirable end, and as such she must be
tolerated.
"Come down this walk with me," she said, in a low tone; "come
quickly, before those girls see us. I want to say a word to you." She
took Biddy's hand as she spoke and hurried her into a little sheltered
path which led round to the back of the house.
"Now, Bridget," she said, "I must clearly understand how matters
are going to be. Dorothy Collingwood cares nothing at all for you;
she is a most fickle girl. She took you up to a certain extent when
first you came, but her conduct during your punishment proves how
little she really cares for you. She and Evelyn will be all in all to each
other, and if you go back to them, you will soon see for yourself that
three is trumpery; now, on the other hand, if you will be guided by
me, I will keep my promise to you. I am willing to become your
chum, and if I am your chum, I will see you safely past all the rocks
ahead. You know nothing whatever about school. There are two
sorts of girls at every school; there is the girl who is always in
trouble, who doesn't learn her lessons, who doesn't obey the rules.
Such a girl is a misery both to herself and her companions. There is
also the girl who obeys the rules, and who learns her lessons. I
represent the one sort of girl, you represent the other. I can teach
you to become like me, without making things at all unpleasant to
you, but you must choose at once; you must be on my side, or on
Evelyn Percival's side. Now which is it to be?"
"Yours, of course," said Bridget; "you are the only girl in the school
who was kind to me to-day, so of course I'll be on your side."
"Very well, that's all right. You must copy me when you talk to
Evelyn Percival. You must show Dorothy also that you resent her
coldness. There's the tea gong. Let us go in. Immediately after tea
you will find time to write that letter to your father, won't you, dear?"
"Yes, of course. I know he'll give me as much money as I want."
"Ask him for plenty; there's nothing like money when all is said and
done. Now come along to tea. I won't be able to sit near you,
Bridget, but I'll have my eye on you, so don't forget how I'll expect
you to behave."
 CHAPTER X.
CHECKMATE.

There was great astonishment among the girls who met at the
Lookout the next day when Janet pronounced in calm, decided tones
that a new member was willing to join the committee, that the new
member was the Irish girl, Bridget O'Hara, who would help her at
her stall, and would give as much money to the cause as was
necessary to insure its success.
"Bridget O'Hara is not here," said Janet, "but she has asked me to
speak for her. She has written to her father to ask him to send her
plenty of funds. She will be more or less of a cipher, of course, but
having the wherewithal she will be a useful one. I propose,
therefore," continued Janet May, "that our committee remains as it is
with this one welcome addition, and that Evelyn Percival is not asked
to join."
While Janet was speaking Dorothy's rosy face turned very pale.
"Now I understand," she murmured; "now I can account for poor
Biddy's change of manner. O Janet, why didn't you leave her alone?"
"What do you mean?" said Janet, flashing round angrily. "Bridget's
help is most desirable. She has money, and she won't interfere with
projects already formed. Had Miss Percival been asked to join, she
would, of course, have given us plenty of money, but she would also
have interfered. I may as well plainly say that I don't choose to be
interfered with at this juncture. That is plain English, I hope; you can
make the worst of it, girls, all of you! I prefer that poor nonentity of
a Bridget to Miss Percival, and I have managed to have my way."
"I suppose we must vote for Biddy," said Ruth and Olive.
"Of course, you must vote for her," retorted Janet.
"I do not object to her joining the committee," said Frances; "but I
think you have managed the whole thing in a very underhand way,
Janet. You are fond of saying that you like frank opinions, so there is
mine for you."
"All right!" said Janet; "I accept it for what it is worth. Now then,
girls, this weighty matter is settled. Dorothy, you must say
something nice to Evelyn. Of course, you have a reasonable excuse
to give her. It would be ridiculous to ask her to join us at the
eleventh hour. She is a sensible girl, and will——"
At this moment, Olive, who was bending over the parapet, turned
round, and said to her companions in a low, almost awestruck voice:
"Mrs. Freeman is coming up the steps of the Lookout!"
The next instant the smiling face of the head mistress appeared.
"Well, my dears," she said, "I won't waste your valuable time a
single moment longer than is necessary. I am very much pleased
with all your zeal in getting up this little bazaar. I, on my part, will
take every possible pains to see that your Fancy Fair is well
attended. I have a suggestion, however, to make; it is this: Evelyn
Percival ought to be asked to take a prominent part in the
management of the fair. She has come back in sufficient time for
this; her health is quite restored, and it is due to her position in the
school to pay her this respect. I dare say, my loves," continued Mrs.
Freeman, "that you have all thought of this already, and are even
now preparing to ask her to join you. If so, you will find her in the
summerhouse at the end of the East Walk with Kitty Thompson.
Good-by, my dears! Forgive me if I have interfered unnecessarily."
Mrs. Freeman went away. The girls had no time to ask her a
question. The head mistress was always quick and decisive in her
movements. She was kind, even indulgent, but she was also firm.
From Mrs. Freeman's decision each girl in the school felt there was
no appeal.
As her retreating footsteps sounded on the winding stairs of the little
tower, the girls who formed the committee for the Fancy Fair looked
at one another. In Janet's gaze there were open-eyed consternation
and dismay. Olive and Ruth appeared what they were: the very
essence of uncertainty and nervousness. Frances Murray could not
restrain an expression of triumph appearing in her bright eyes, while
Dolly looked both glad and sorry.
"O Janet!" she said, "I wish I could take your side and my own. I
wish I could obey dear Mrs. Freeman, and have our darling Evelyn to
help us, and be one of us, and I also wish to do the thing that
makes you happy."
"Oh, don't worry about me," said Janet. "Of course, the thing is
inevitable. Under existing circumstances, I give in. I have only one
request to make, girls, and that is, that you will not betray to Evelyn
Percival, who, of course, will take the lead now in the management
of the Fancy Fair, the very frank objections I have made to having
her with us. We must welcome her, of course, with a good grace,
and I trust to you all to keep my little remarks to yourselves."
"Of course, of course, Janey," they each eagerly replied.
"As if we could be so mean as to tell," remarked Ruth, going up to
her friend and giving her hand a squeeze.
Janet did not return the pressure of Ruth's hand. She turned
abruptly to Dorothy.
"Evelyn is to be found in the summerhouse. Will you go and fetch
her at once, Dolly?"
Dorothy ran off without another word. While she was absent Janet
kept her back to her friends. She generally carried a little sketchbook
in her pocket; she took it out now, and under the shelter of her
parasol pretended to sketch the lovely summer landscape which
surrounded her.
The other girls who were watching saw, however, that her small,
dainty fingers scarcely moved.
When voices and steps were heard in the distance, Janet was the
first to turn round, and when Evelyn appeared on the scene Janet
went up and bade her welcome.
"We have elected you to join our committee," she said, in a low and
careless voice. "As the head girl of the school, you will naturally take
the lead in the matter; but, as you have been obliged to be absent
when our scheme was first started, you would perhaps like me to tell
you how far we have gone."
"I am delighted to join the committee," replied Evelyn, "and
particularly glad that you have asked me, Janet. You may be sure,
girls, I'll do all I can to help, but as the idea of the Fancy Fair was
yours, Janet, I don't think I ought to take the lead."
For a second a pleased expression flitted across Janet May's cold,
self-possessed face. It vanished, however, as quickly as it came.
"No," she said, "I cannot possibly take the lead. The head girl of the
school has certain rights which no one must deprive her of. It is
generous of you to offer me your place, Evelyn, but, even if I
allowed myself to accept the position, Mrs. Freeman would instantly
require me to vacate it in your favor. The thing is settled, then; you
are formally invited by us all to join our committee; is that not so,
girls?"
"Yes, yes," they all exclaimed, delight and relief plainly apparent on
every face.
"You are formally elected, therefore," proceeded Janet. "Won't you
sit down, Evelyn? That is a comfortable seat in the shade over there.
Won't you take it? I can then tell you as briefly as possible what we
have done."
Evelyn sat down in the comfortable seat without a word. Frances
Murray sprang to her side, slipped her hand through her arm, and
looked into her face with adoration; Ruth and Olive were only
restrained by Janet's presence from groveling at her feet. Dolly alone
leant in a careless attitude against the low parapet of the tower. Her
affectionate glance traveled many times to her friend's face, but she
had too much tact and too good taste to show her preference too
openly while Janet May was present.
"Up to the present," said Janet, also leaning against the parapet,
and exactly facing Evelyn, "up to the present I have managed the
proposed bazaar. If it is generally wished, I can still remain treasurer.
At the present moment, I am sorry to say, there is very little money
to guard. If the thing is to be a success, more money must be spent,
but that, of course, is for Evelyn to decide. We are having the
bazaar, Evelyn, hoping to raise money to send little Tim Donovan to
a good school. Mrs. Freeman said something about this bazaar being
repeated, if necessary, in the future; but that, of course, we need
not discuss at present. The bazaar is to be called a Fancy Fair. It will
be held in a large tent in the four-acre field. This part of the
entertainment Mrs. Freeman has herself promised to provide. Our
present idea is to have four stalls. You will, of course, conduct the
principal one; I, if permitted, will take the second; Dorothy or
Frances Murray will manage the third; and there will also be a
refreshment stall, for which we have not at present provided. Each
girl of the committee has undertaken to secure a certain number of
fancy materials for sale at the fair. Ruth, Olive, and I at the present
time are doing well; about six little girls of the lower school are
helping us. We meet twice a week in the summerhouse at the end of
the South Walk to work for the bazaar, and the results will, I believe,
be fairly creditable. I cannot say what arrangements Frances is
making, but she will doubtless tell you herself. Dorothy is also the
soul of industry. You'll probably reconstruct everything, and I shall
be ready to come to you for advice whenever you ask me. There is,
I think, only one thing more to say, and that is, that I have
persuaded the new girl, Bridget O'Hara, to join us. She does not
strictly belong either to the upper or the lower school at present. Her
position in the house is, I think, somewhat unique. She is a very tall,
grown-up-looking girl, but she is not yet quite fifteen years of age.
Her mind very much resembles her body, being extremely grown-up
in some ways, and absolutely childish in others. Her acquirements
are also those of a child. I have thought it right, however, in your
absence, of course, Evelyn, to ask her to join us. She has a good
deal of originality; she has also some money, which she is willing to
devote to the cause. I think that is all. I am now going to join my
workers in the summerhouse at the end of the South Walk. You,
Ruth, and you, Olive, can come with me if you like, but if you prefer
it, you are quite at liberty to join Evelyn's stall, for now that I have
got Bridget's help I can do admirably without you."
Ruth and Olive looked more undecided than ever, but Evelyn said in
a firm voice: "Of course, girls, you could not for a moment wish to
desert Janet. I should like to say one thing before you go, Janet; it is
this, that I am very much surprised at your pluck and bravery in
getting up a bazaar of this sort. I am pleased to join it, and to do all
I can to promote it. Under the circumstances, I should much prefer
working as your aide-de-camp to taking the lead; but you are quite
right in saying that the head girl of the school has certain privileges
which, whether she likes it or not, she cannot forego. I must, of
course, take the principal part at the bazaar, but I shall, in every way
in my power, do what is most agreeable to you, and will lose no
opportunity to let my friends know that the idea is yours, not mine."
"You are very good-natured," said Janet, "but I, too, have something
to say. Under the circumstances, I prefer sinking into the
background. After all, the only person to be seriously considered is
little Tim Donovan. If he is substantially helped I don't suppose it
matters much what anyone thinks of us."
 CHAPTER XI.
A WILD IRISH PRINCESS.

The girls of the lower school were all busy with their preparation.
Violet and Rose sat side by side. They had been chums for nearly a
year now, and the fact was so fully recognized in the school that
even their desks were placed close together. Violet was puzzling her
little brains over a very difficult piece of French translation, Rose
endeavoring to learn four or five long stanzas from Scott's "Lady of
the Lake." They were both clever little girls, and, as a rule, their
preparation was quickly over, and their tasks speedily conquered;
but to-night there was a holiday feeling in the air; a sense of
idleness pervaded everyone. Lessons seemed cruel, and the children
rebelled against their tasks. They looked at one another, laughed,
yawned, struggled with the listlessness which seized them, shot
envious glances at their more studious companions, and absolutely
refused to overcome the difficulties of the French translation and the
English poetry.
The door between the lower schoolroom and the room occupied by
the girls of the middle school had been thrown open, and from
where the children sat they could see the pretty flounce of a pale
blue muslin dress, and the provoking and exasperating peep of a
little, pointed, blue Morocco shoe. The shoe evidently belonged to a
restless foot, for it often appeared beneath the flounce, to vanish as
quickly, and then to poke itself into notice again.
"It's Biddy," whispered Violet in a low tone to Rose. "I don't believe
she's learning her lessons a bit better than we are."
"She never learns them at all," answered Rose. "Janet does them for
her now; don't you know that, Violet?"
"Hush!" said Violet, "we are disturbing Katie and Susy Martin, and
they are such spiteful little cats, they are sure to tell on us. Hush! do
hush, Rose! you ought not to say such things."
"I won't say them if you don't like," whispered Rose back again; "but
they are true all the same."
Violet bent over her French translation. Rose made another frantic
struggle to conquer "The Lady of the Lake."
The other children in the room were working with considerable
industry; the little idlers in the corner had to suppress their emotions
as best they could.
Rose had a very emphatic way; she was a stronger character than
Violet, and in consequence had her little friend more or less under
her thumb.
Violet had a great admiration for Biddy, and, as she was really an
honorable and conscientious child, Rose's words shocked her very
much.
The moments went by. The summer evening outside looked more
beautiful and inviting each moment. After preparation was over,
there was a treat in store for the children. This was Bridget O'Hara's
birthday, and she was herself the giver of the treat. The children
were to have a sort of supper-tea in the tent on the lawn, and
afterward Biddy was going to give each of them a little present in
memory of the day.
The thought of Biddy's present and Biddy's treat had filled every
little heart with a pleasant sense of excitement during the entire
day; but Violet felt now that if Rose's words were really true she
would not care to accept a keepsake from Bridget.
As she sat before her desk, too lazy, too languid, and at the same
time too excited, to pay the smallest heed to her lessons, she could
not help wishing that she could see something more of the blue
frock than just that part which covered the pretty foot.
She slipped down lower and lower by her desk, and presently
contrived to get a view of Bridget's desk. She could not see her face,
but she could catch a glance of a plump young hand; it was quite
still, it did not move, it did not turn a page. Violet could stand it no
longer. In a moment of desperation she kicked off her slipper, and
springing from her seat, bent low on the floor to pick it up.
From there she could see the whole of Biddy's figure. Oh, horror!
her little heart went down to zero; Bridget O'Hara's head rested
against her plump hand; she was fast asleep.
The shrill voice of mademoiselle was heard from her corner of the
room:
"Reste tranquille, mon enfant; tu es bien ennuyeuse; est ce que tu
ne sais pas que c'est l'heure de silence?"
Violet scrambled to, her feet, and sat down before her French
translation with a crimson face.
In the meanwhile a pale, quiet-looking girl had entered the room
where the middle school were busy over their tasks, and, bending
down by Bridget O'Hara's side, took up an exercise she had just
finished, and looked over it swiftly and eagerly.
"That is right," she said; "you will get good marks for this. Now,
what about your arithmetic?"
"I have managed my sums fairly well, Janet; see," pulling an
exercise-book forward. "I suppose they are all right, but they look
very funny."
"They must be all right, dear. Let me see! Yes, yes; oh, what an
incorrigibly stupid girl you are! This sum in compound subtraction
has got the answer which should be attached to the compound
addition sum. Quick, Bridget, give me your pen; I will score through
these two lines, and then you must add the figures underneath
yourself. That is right. What have you done with my——"
"Your copy, Janet? I was going to tear it up, as I had done with it."
"Don't do that, give it to me; it will be safest. Now, try and look over
your poetry, Bridget. I will wait for you outside."
"Oh, that is easy enough; I shan't be any time. It's the first page or
two of that delightful 'Ancient Mariner'; I can get it done in no time."
"Lucky for you. I will wait for you outside; I have something I want
to say to you. Be quick, for all those small tots will be out
immediately, and they'll want to take up every moment of your time.
Give me those notes, however, before I go."
Bridget pulled some crumpled bits of paper out of her pocket, and
thrust them into Janet's eager hand.
Miss May left the room, and Biddy, wide awake now, devoted herself
to her poetry.
There was an eager, pleased, almost satisfied, expression on her
face.
It was over a week now since Janet had taken her up. During that
time she had, without in the least guessing the fact herself, been
brought into a considerable state of discipline.
If she obeyed no one else in the school, Janet's slightest nod was
sufficient for her.
It was Janet's present aim, whether by foul means or fair, to make
Biddy appear both good and fascinating.
She did not want her captive to feel the end of her chain; she was
clever enough to make Biddy her complete slave without allowing
the slave to be conscious of her slavery.
The result of this week of very judicious slavery was, as far as
externals went, highly beneficial.
Biddy had a gorgeous taste in the matter of dress. She wore her
splendid garments with truly barbarian recklessness, overdressing
herself on one occasion, being untidy and almost slovenly on
another. A few suggestions, however, from Janet, altered all this,
and the most fastidious person could now see nothing to object to in
the clothes which adorned her beautifully proportioned figure, and
the hats under which that charming and lovely face looked out.
To-night, Biddy's pale blue muslin, made simply, but with a lavish
disregard to expense in the matter of lace and ribbons, was all that
was appropriate; her crisp chestnut curls surrounded her fair face
like a halo. There was a queer mixture of the woman and the child
about her; she was by many degrees the most striking-looking girl in
the school.
It took Biddy but a very few minutes to conquer the difficulties of
"The Ancient Mariner." She had a great aptitude for committing
poetry to memory, and after repeating the stanzas two or three
times under her breath, she slipped the book inside her desk and ran
out.
To do this she had to go through the schoolroom where the little
girls, Violet and Alice, were sitting mournfully in front of their
unlearned lessons.
"Oh, you poor tots!" she said, struck by the expression on their
wistful faces, "haven't you done yet? The feast is almost ready. I've
ordered clothes baskets of strawberries, my dears, and quarts and
quarts of cream."
"Silence, mademoiselle!" screamed the French teacher.
Bridget put her rosy fingers to her lips in mock solemnity, blew a kiss
to all the children, and banged the door somewhat noisily behind
her.
Violet's blue eyes sought Alice's; there was a world of entreaty in
their meaning. Alice began, with feverish, forced energy, to mutter
to herself:

"A chieftain's daughter seemed the maid."

Violet continued to gaze at her; then, taking up a scrap of paper, she

scribbled on it:

"I don't believe that Janet helps Biddy with her lessons."

This scrap of paper was thrust into Alice's hand, who, in a moment,
tossed a reply into Violet's lap:

"Yes, she does. You ask Honora Stedman or Jessie Sparkes."

Violet tore the paper into a thousand bits. Tears, she could scarcely
tell why, dimmed her pretty eyes. She sank back in her seat, and
resumed her lessons.
"Maintenant, mes enfants, l'heure de préparation est passée," said
the French governess, rising, and speaking with her usual, quick
little scream. "Mettez vos livres de côté; allons-nous à la fête donnée
par la gracieuse Mlle. Bridget O'Hara."
The children jumped up with alacrity. Chairs scraped against the
floor; desks were opened and books deposited therein more quickly
than quietly, and then the whole eager group went out.
There was a large tent erected on the front lawn; gay flags were
posted here and there round it, and a rustic porch had been hastily
contrived at the entrance. This was crowned with many smaller
flags, and was further rendered gay with bunches of wild flowers
and ferns which had been fastened to it, under Bridget's supervision,
early in the day.
The brilliant effect of the many colored flags and banners, the peep
within the tent of tempting tables and many charming presents,
excited the wild spirits of the little ones to an almost alarming
degree.
Alice looked at Violet with a face full of ecstasy.
"How I love Biddy O'Hara!" she exclaimed. "Think of her getting up
such a lovely, exquisite treat for us! Would any other girl think only
of others on her birthday? Oh, I love her; I do love her!"
"But if she does really crib her lessons!" answered Violet, in a low
tone of great sorrow. "O Alice, it can't be true."
"It is true," replied Alice; "but, for goodness' sake, Violet, don't fret
yourself; it isn't our affair if Biddy chooses to do wrong. Whether she
does right or wrong, I shall still maintain that she's a dear, generous
darling. Do come on now, Violet, and let us enjoy ourselves." Alice
caught her little companion's hand as she spoke, and the two
children ran down the rather steep grassy incline to the tent.
Most of their companions had arrived before them, and when they
entered under the flower-crowned porch, they found themselves in
the midst of a very gay and attractive scene. Bridget, with two or
three older girls of the school, was entertaining the children with
strong sweet tea, piles of bread and butter, cakes of various sizes
and shapes, and quantities of strawberries, which were further
supplemented with jugs of rich cream.
Violet and Alice seated themselves at once at one end of the long
table, and the merry feast went on.
What laughter there was at it, what childish jokes, what little
harmless, affectionate, mirthful repartees! Bridget O'Hara's face
wore its sweetest expression. The Irish girl had never looked more in
her element. Frances Murray and Dorothy, who were both helping
her, had never seen Bridget look like this. She showed herself
capable of two things: of giving others the most intense pleasure
and enjoyment, and absolutely forgetting herself.
Dorothy had not felt kindly disposed to Bridget during the past week.
Bridget's conduct, Bridget's extraordinary reserve, the marked way in
which she resented small overtures of friendship from Evelyn
Percival, hurt her feelings a great deal; but to-night Dorothy
Collingwood felt her heart going out to Biddy in a new, unexpected
way.
"I agree with Evelyn," she said suddenly, turning round and speaking
to Frances Murray.
"About what, my dear?" retorted that young lady. "You generally do
agree with Evelyn, you know."
"Don't tease me, Frances; of course we're chums, but I hold, and
always will hold, my own opinions. I agree with her now, however. I
agree with her with regard to Bridget O'Hara."
"Biddy looks very sweet to-night," replied Frances, "but surely Evelyn
cannot care about her."
"Biddy has been very nasty to Evelyn," answered Dolly. "Of course, I
know who is really to blame for it. Still I thought Biddy would have
more spirit than to be led in a matter of this sort. But do you think
Evelyn resents this sort of thing? Not a bit of her. She is just as
sweet and good about it all as she can be, and she said to me, what
I am really inclined to believe, that if Biddy is only done justice to,
there won't be a nobler woman in the world than she."
"Oh, fudge!" said Frances; "I grant that she does look very sweet
now, but it's just like Evelyn to go to the fair with things, and it's just
like you, Dolly, to believe her. Come, come, the little ones cannot eat
another strawberry, however hard they try, and Bridget is going up
to the end of the tent to distribute the presents."
"Let us see," replied Dolly.
The two girls went up to the far end of the tent, where a little table
covered with a crimson cloth stood; on this Bridget had placed her
small gifts.
They were all minute, but all dainty. They had arrived from Paris, a
few nights ago, in a small box. Thimbles in charming little cases,
dainty workboxes, writing cases, penholders, dolls, photograph
frames, boxes of colors, etc., etc., lay in profusion on the pretty
table.
Biddy stood by her presents, a bright light in her eyes, a bright color
on her cheeks. The two elder girls, who stood in the background,
could not help a sudden pang as they watched her. There was
something about her mien and bearing which made them, for the
first time, clearly understand that this girl was a wild Irish princess
at home. For the first time they got an insight into Biddy's somewhat
complex character.
"Come here, darlings," she said to the children in her sweet, rather
low-pitched voice. "I am glad to give you a little bit of pleasure. It is
the best sort of thing that can happen to me, now that I'm away
from father. Had you enough to eat, pets?"
"Oh, yes, Biddy, oh, yes!" they all cried.
"That's right. I thought you would. We have lots of feasts of this sort
at the Castle. The children aren't like you, of course; they live, half
of them, down in the cabins near the water's edge, and they come
up with their little bare feet, and their curly heads that have never
known hat nor bonnet, and their eyes as blue as a bit of the sky, or
as black as the sloes in the hedges. Oh, they are pets every one of
them, with their soft voices, and their little prim courtesies, and their
'Thank you, kind lady,' and their 'Indeed, then, it's thrue for ye, that
I'm moighty honored by ateing in the sight of yer honor.' Ah, I can
hear them now, the pets! and don't they like the presents afterward,
and don't they send up three cheers for father and me before they
go away! They are all having a feast to-night at the Castle in honor
of my birthday, and father is there, and all the dogs, but I'm away; I
expect they're a bit lonesome, poor dears, without Biddy, but never
mind! You have all been very good to let me give you a little feast,
my dear darling pets."
There was a great pathos in Biddy's words; the children felt more
inclined to cry than to laugh; Dolly felt a lump in her throat, and
even Frances looked down on the ground for a second, but when
there was a brief pause Frances raised her hand, and waved it
slightly as a signal.
This was enough, all the hands were raised, all the handkerchiefs
waved, and from every throat there rose a "Hip! hip! hurrah!" and
"Three cheers for the Irish princess!"
"Many happy returns of the day," said Frances, and then all the
children repeated her words.
"You must not add any more," exclaimed Biddy. "I don't wish to cry;
I want to be happy, as I ought to be when you are all so nice and
good to me. I may as well say frankly that I did not at all like school
at first, but I do now. If you are all affectionate and loving, and if
Janet goes on being kind to me, I shall like school, and I shan't mind
so much being broken in."
"Poor Biddy," exclaimed Dorothy, turning to her companion; "she
reminds me of the lovely silver-winged horse Pegasus. She does not
like the taming process."
"No, my dear, that's true," replied Frances; "but Pegasus grew very
fond of Bellerophon in the end."
"Only I deny," said Dolly, "that Janet is in the least like Bellerophon."
"Listen!" exclaimed Frances.
"I am going to give you your presents now," said Bridget. "Come
here, each of you in turn."
The children pressed eagerly to the front, and Biddy put a small gift
into each of their hands.
"Now come for a walk with me," she said. "I shall tell you a fairy
story—a very short one; it pleased the barefooted children at home,
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebooknice.com