Project MODEL 28

Dataset Description

Our group will be using the Intel Image Classification Dataset, a publicly available Kaggle collection of approximately 24,000 RGB images at a uniform 150×150‐pixel resolution. The data are organized into six classes‐named subdirectories—buildings, forest, glacier, mountain, sea, and street—each assigned integer labels 0 through 5. The original download provides separate seg_train and seg_test folders (roughly 14,000 training and 3,000 testing images), and we will further split the training set 80/20 for validation. All files are in JPEG format, balanced across classes, and require no additional annotation, making this dataset straightforward to load, preprocess, and feed into our flattened‐pixel (67,500‐feature) and CNN pipelines.

Project Demo

We have demonstrated the working of the project as well as a walkthrough on how to run the web application on your own development machine once cloned!
-> Intel Classifier Dataset Project Demo

Clone the repo

Use the https or ssh link to clone the repo to train the models or test the models that we have already trained.

⚠️ Git LFS Notice: Large Model Files

This project uses Git Large File Storage (LFS) to manage large model files such as:

• best_mlp.h5
• best_model.keras
• classes.npy
• train_logreg.h5

These files are required for the Flask demos in the /flask_deploy/ directory.

⚠️ Before cloning or pulling this repository, please make sure Git LFS is installed on your system.

✅ To install Git LFS:

macOS (with Homebrew):

brew install git-lfs
git lfs install

Ubuntu/Debian:

sudo apt update
sudo apt install git-lfs
git lfs install

Windows:
Download and install from https://git-lfs.github.com
Then clone repository as usual:

git clone https://github.com/yourusername/your-repo.git

Git LFS will automatically fetch the required model files when you clone or pull.

Data Preprocessing

Before training any model, you need to preprocess the raw image data. Follow the steps below to generate ready-to-use .npy files for your assignments.

---

Download the data from Kaggle

Download the Intel Image Classification Dataset from Kaggle:
-> https://www.kaggle.com/datasets/puneet6060/intel-image-classification

After downloading:

• Unzip the dataset
• You will get a folder named archive/
• Move the entire archive/ folder into the root directory of this project

---

1: Go to the project root directory

Open your terminal and make sure you are in the root directory of the project.
To verify, run:

ls

If you see the following folders and files:

archive/  data_preprocess/  figures/  models/  notebooks/  README.md

then you are in the correct place.

---

2. Run the preprocessing script

Run this command:

python data_preprocess/preprocess.py

This will:

• Traverse all class-named folders under archive/seg_train/seg_train
• Resize every image to 150×150
• Normalize pixel values to the range [0, 1]
• Flatten each image into a 1D array
• Save two output files in the current directory:
  • X.npy — the image data
  • y.npy — the corresponding labels

---

What are X.npy and y.npy?

File	Shape	Contents	Used for
X.npy	(N, 150×150×3) → (N, 67500)	Flattened, normalized RGB image data	Input features for MLP, LogReg, CNN
y.npy	(N,)	Integer class labels (e.g., 0–5)	Target labels for classification

These arrays will be the basis for your exploratory data analysis (EDA), training, validation, and testing.

---

3. Load the .npy files in your script

Once you've run the command above, you can load the data in any script like this:

import numpy as np

# load preprocessed features and labels
X = np.load("X.npy")
y = np.load("y.npy")

print("X shape:", X.shape)   # (N, 67500)
print("y shape:", y.shape)   # (N,)

• You can now use X and y directly in your:
  • Logistic Regression models
  • Multi-Layer Perceptrons (MLP)
  • Convolutional Neural Networks (CNN) (after reshaping)
  • EDA visualizations
  • Train/test splits and k-fold cross-validation

---

Once this step is complete, you're ready to move on to model building and analysis.

Testing the model

**We will use flask to run and test our models. ** So install flask in your preferred virtual environment using the following command:

conda install anaconda::flask

Testing Logistic Regression Model:

For logistic regression,
You can train the model in the logistic_regression.ipynb and get the weights in intel_clf.joblib and intel_le.joblib.

Next,
flask_logreg.py has the flask implementation and testing. templates/index.html has the implementation of the website. For now, there is only logistic regression model available. Go to models/logreg/ and run the code below to check it out:

python flask_logreg.py

Testing Multi-Layered Perceptron Model:

For MLP,
You will have to train the model and create a best_mlp.h5 file to save your model.
For doing so, go to models/mlp and run:

python train_mlp_taylor_submission_non_ipynb.py

Next,
templates/index.html has the implementation of the website. To run the website + flask, go to models/mlp and run the code below for html+flask testing or demo environment:

python flask_mlp.py

Testing CNN Model:

For testing CNN, You will have to train the model and create a best_model.keras file to save your model. For doing so, open the cnn_training.ipynb under models/cnn/ and run all the cells.

Now we have a saved model. Then make sure you are in models/cnn directory and run the following:

python cnn_flask.py

Testing the project

For testing the project, go to flask_deploy directory and you will see an app.py file.
Then run:

python app.py

This will provide you with a link http://127.0.0.1:3000 or similar.
Click on that link, and you will be able to test all three model predictions for the image you uploaded.