Implementing a Basic Fully Connected Neural Network in Flask Based on the Iris Dataset

This is the second project for the DSS course (CIS431) at JUST University (2^nd semester, 2023).

It contains the guidelines and code necessary to help the students to carry out the project.

The following datasets have been used (for demonstration purposes):

• Iris, the famous iris dataset; will be used for building an FCNN classifier, which will be deployed on Flask.
• Animals-10, a dataset which contains 10 types/classes of animals in images; it will be used as an object recognition task for building a CNN architecture.

Note: The provided code based on these datasets is a mere reference for you; you are required to build and deploy a model based on another dataset (discussed in the objectives section).

Update (19/5/2023)

There is a missing layer in cnn_animals.py that might cause an ambiguity on what the input dimension is when loading the model; to mitigate this issue, you will have to add this line as the first layer in the neural network: Input(shape=(None, None, 3)).

As for predicting an already trained model, you will have to expand the dimensionality of the sample image before invoking the predict method; you can do this by calling np.expand_dims(your_image, axis=0).

Getting Started

Clone the project from GitHub

$ git clone https://github.com/tariqshaban/fcnn-iris-flask.git

Make sure that you have Python installed.

No further configuration is required.

Note: If you do not know how to clone a repository, you can either directly download the project or read here.

Warning: You must install the packages in the exact version found in the requirements.txt file; you can use this command: pip install -r requirements.txt.

Project Structure

├── README.md                                                 <- README file for developers using this project
├── requirements.txt                                          <- Contains the dependencies required to successfully run the project
│
├── dataset
│   │── animals-10                                            <- Store the dataset images of the animals-10
│   │── synthetic-asl-numbers                                 <- Store the dataset images of the synthetic-asl-numbers (your target dataset)
│   └── iris.csv                                              <- Store the iris dataset
│
├── out                                                       <- Store model weights/configuration/architecture/...
│
├── static
│   │── images                                                <- Store images for display in the frontend
│   │── scripts                                               <- Store Javascript for execution in the frontend
│   └── styles                                                <- Store CSS for styling the frontend
│
├── templates                                                 <- Store HTML files which are linked to Flask
│
├── backend.py                                                <- Handles Flask endpoints
├── cnn_animals.py                                            <- Handles the training of the animals-10 dataset using CNN
└── fcnn_iris.py                                              <- Handles the training of the iris dataset using FCNN

Objectives

Now that you have been provided with an example of how to build, export, and deploy a model, your task will be composed of:

• Inspect the synthetic-asl-numbers dataset; note that this repository contains a modified version of the dataset. Please work ONLY on the dataset in the repository and do not download the dataset from Kaggle.
• Building a Convolutional Neural Network (CNN or ConvNet)
  • Use the dataset in dataset/synthetic-asl-numbers/ as the target dataset.
  • Use the cnn_animals.py script and modify:
    • The dataset source (make it based on synthetic-asl-numbers)
    • The dataset labels
    • The model's architecture
    • Any fine-tuning (as necessary)
• Export the trained model
• Deploy the exported model on Flask
  • Frontend
    • Create a home page, which contains a form having:
      • Upload button (for images only)
      • Submit button
    • Create a prediction page; which displays the prediction to the user.
  • Backend
    • Add a default endpoint: /, which redirects the user to the home page.
    • Add a prediction endpoint: /predict, which does the following:
      • Properly read the uploaded image.
      • Pass the image into the model for prediction.
      • Fetch the prediction of the model.
      • Redirect the user to the prediction page while passing the prediction.
    • You can copy the backend.py script and modify it accordingly based on the given requirements.
• Verify your work by uploading a test image.

Submission

You will need to submit the following files:

• The code used for building and exporting the CNN model.
• Basic performance indicators for your CNN model, which include:
  • Loss history plot.
  • Accuracy history plot.
  • Confusion plot.
  • *These plots are automatically exported to you as an image file in cnn_animals.py.
• The code used for deploying the CNN model (using Flask), which includes:
  • The front end used as an interface for the user, including, but not limited to:
    • Template folder (HTML files).
    • Static folder (JS, CSS, and raw images used as an asset).
  • The backend is used for processing the user's requests.

You should upload all these files to the e-learning; however, the Flask code should be on Replit. Upload a shareable link of that project to the e-learning.

Rubric

The following points will dictate your grading:

Criterion	Weight
CNN model is trained and exported	1 mark
CNN model returns acceptable loss/accuracy values	1 mark
The end-user can upload an image	1 mark
The backend can interpret "decode" the uploaded image	1 mark
The model returns a valid prediction	1 mark
The backend successfully redirects the prediction and displays it to the user	1 mark
The displayed prediction is in a textual form rather than numerical	1 mark
The website's visuals are appealing	1 mark
The web application is hosted on Replit	1 mark
The student understands how to construct a neural network	1 mark

Notes

Note: Do not return the numerical value (1, 2, 3) of the prediction directly to the user; instead, return a textual value (one, two, three).

Note: Your target dataset is the synthetic-asl-numbers, DO NOT BUILD A MODEL WHICH IS BASED ON ANOTHER DATASET.

Note: You are not allowed to add, modify, or delete the target dataset in any way.

Note: If your computational resources are limited (very long training times), use Colab instead of your local machine.

Note: You are expected to deliver a properly formatted website; this includes structuring and styling of the webpages.

Warning: Any form of cheating will result in zero for both parties.

Warning: If you were not able to defend your code or understand the general intuition of the project, you will get zero.

Warning: Late submissions are not allowed. No exceptions!

Useful Links

CNN intuition (IBM topic)

Minimal CNN Code using Keras (Official TensorFlow Docs)

Official Flask Docs

Creating a Flask project on Replit

Rand's Video

Rand's Replit Project

---