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

Implimented a simple Gradio interface to infer on CustomResNet model and get GradCAM results

Task :

Classification on CIFAR10 dataset using Custom ResNet model by using pytorch lightning.

Files :

-> requirements.txt file contains necessary packages to install.

-> custom_resnet.py file contains model architecture.

-> CustomResNet.pth contains trained model checkpoints (weights).

-> examples folder : 10 example images like cat.jpg, car.jpg,..

--> app.py contains gradio code. By using gradio here implemented by selecting input images or examples output display the gradcam image and prediction and top k classes.

--> misclassified_images folder : 10 misclassified images

Implimentation :

First loaded the model by using model weights .pth file.

By using GRADIO we created these features :

-> Asking the user they want to see GradCAM images if yes then how many images, from which layer and also allow opacity change.

-> Providing the option to user they want to view misclassified images, and how many images. If they want to apply grad cam for misclassified images.

--> Option to upload new images, as well as select from 10 example images.

--> Providing one more option how many top classes they want to see.