FULL STACK ML

---

This web app allows the user to upload a photo of a dog or a cat and get the prediction as to whether there is a dog or a cat in the photo

📝 Table of Contents

About | The Dataset | Building a CNN model with TensorFlow | Front end | Back end | Web app URL and usage | Authors | Conclusions

About

The purpose of this personal project is, first and foremost, to strengthen the knowledge in machine learning and NN in particular. This machine knows how to take a picture from a user and make a prediction as to whether the image is of a dog or a cat.

This project implements several technologies: I build the model with TensorFlow using CNN

On the client side, I used HTML JS and TensorFlow JS.

On the server side, I used NODEJS and IBM WATSON MACHIN LEARNING to host the model.

Below is the project architecture:

The Dataset

The data set I used is from KAGGLE, and you can check it here

The data set contains 25K images of dogs and cats divided into two classes (dog/cat). The photos come in different sizes, so it is crucial to perform pre-processing and correct the size to match the model's requirements. Each image consists of 3 layers of RGB color.

Some Random images of the Dataset

Building a CNN model with TensorFlow

The model was built in Python in the Jupiter notebook environment

Our dependence:

Numpy
Tensorflow

1. Preprocessing: The first step is a preliminary inspection of the images to find corrupted pictures and delete them or convert them to an image that can work with TF.

2. Upload the images: I used tf.keras.utils.image_dataset_from_directory Generates a tf.data.Dataset from image files in a directory.

3. Splitting into Train Validation and Test Datasets and check the class balance in each dataset

4. Build the model- a CNN model

model = tf.keras.Sequential([
    tf.keras.layers.Rescaling(scale=1./255, input_shape=(128,128,3)),
    
     
    tf.keras.layers.Conv2D(32, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(),
     tf.keras.layers.Dropout(0.25),
    
    tf.keras.layers.Conv2D(64, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(),
     tf.keras.layers.Dropout(0.25),
    
    tf.keras.layers.Conv2D(128, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(),
     tf.keras.layers.Dropout(0.25),
    
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(512, activation='relu'),
     tf.keras.layers.Dropout(0.5),
    tf.keras.layers.Dense(2, activation='softmax' )
])

with Adam optimizer and loss function SparseCategoricalCrossentropy

5. Fit the model with 20 epochs and evaluate the model.

• The evaluation gave me 83 percent accuracy, not perfect, but at this point, it satisfies me.

6. deploy the model using IBM Watson machine learning -

I used ibm doc for uploading a scikit model , and I adjusted to my needs

Front end

• USER: The user have the option of uploading a photo from a local device or copying a dog or cat photo URL link ==>

• In the browser: After the user uploads an image, I used TensorFlow js to pre-process the image to match the model inputs. The steps were:

  1. resizing
  2. converting the image to an array.

==> In the next step, the array is sent to the server

Back end

• Using Node Js and express js - I built a server. The server performs two functions:

1. Creating an endpoint for the site itself.
2. Building an API that receives the information from the client and sends it to the IBM Watson API

• in the next stage: The data goes to the model we built with Tensorflow hosted at IBM Watson, then we get a prediction response of whether the image is a dog or a cat, from the server, the data returns to the client and shows him the prediction.

WebApp URL

As of now, the site is live, and a prediction can be made Check it

• Since the goal of the project is to create an end-to-end application, the level of importance of the UI is a low priority, and therefore the design is a simple BOOTSTRAP design, not too flashy but meets the needs of the project

⛏️ Built Using

• TensorFlow - ML Framework for python
• Ibm Watson machine learning - ML model hosting
• Express - Server Framework
• NodeJs - Server Environment

✍️ Authors

• @amit9021 - Idea & Initial work

Conclusions

• My main goal in this project was to specialize as much as possible in the capabilities of Tensorflow. In addition, it was important to me to allow an end user to make predictions independently without the need for coding knowledge. Also, during the project, I often encountered security problems and intensive utilization of the user's resources, so I decided to create the possibility to perform the prediction On the server side and not on the client side.

• The project is not perfect, and those who try to upload another image that is not a dog or a cat will probably be disappointed to see that the model will still return a prediction. Due to lack of time, I leave this glitch (very simple to fix) for fun and invite you to predict whether a person is a dog or a cat