README.md

metadata

title: ERA S12
emoji: 🌖
colorFrom: green
colorTo: purple
sdk: gradio
sdk_version: 3.39.0
app_file: app.py
pinned: false
license: mit

CustomResNet with GradCAM - Interactive Interface

This project Impliments a simple Gradio interface to perform inference on CustomResNet model and generate the GradCAM visualization results.

Task :

The task involves performing classification on the CIFAR-10 dataset using the Custom ResNet model built with PyTorch and PyTorch Lightning.

Files :

1. requirements.txt: Contains the necessary packages required for installation.
2. custom_resnet.py: Contains the CustomResNet model architecture.
3. CustomResNet.pth: Trained model checkpoint file containing model weights.
4. examples/: Folder containing example images (e.g., cat.jpg, car.jpg, etc.).
5. app.py: Contains the Gradio code for the interactive interface. Users can select input images or examples and view GradCAM images, predictions, and top-k classes.
6. misclassified_images/: Folder containing misclassified images.

Implementation

The following features are implemented using Gradio:

1. GradCAM Images: Users are prompted to choose whether they want to view GradCAM images. They can specify the number of images, the target layer, and adjust opacity.
2. Misclassified Images: Users have the option to view misclassified images and apply GradCAM visualization to them.
3. Upload and Select Images: Users can upload new images or select from a set of 10 example images.
4. Top Classes: Users can choose how many top classes they want to see in the prediction results.

Usage

1. Run the app.py script to launch the interactive Gradio interface.