U-Net Autoencoder (Cats vs Dogs) + Grad-CAM

This repo contains a local (non-Colab) notebook that:

• Trains a U-Net CNN autoencoder to reconstruct cat/dog images.
• Uses MSE loss and reports PSNR on a validation split.
• Visualizes Grad-CAM on an intermediate feature map and saves: input, reconstruction, heatmap, overlay.

Dataset

The dataset is included in this project:

• cats_set/
• dogs_set/

The notebook expects to be run from the repo folder (so DATA_ROOT = Path.cwd() works).

Setup

Create/activate an environment (recommended) and install deps:

python3 -m pip install -r requirements.txt

If you use Jupyter, make sure the notebook kernel points to the same Python where you installed packages.

Run

• Open unet_autoencoder_cats_dogs_colab.ipynb
• Run cells top-to-bottom

Outputs are written to outputs/ (not committed).

Deliverables / Links

• Notebook: unet_autoencoder_cats_dogs_colab.ipynb
• Report: included at the bottom of the notebook

After pushing to GitHub, paste your GitHub repository link here.