README.md

metadata

title: S13
emoji: 🐨
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 4.28.0
app_file: app.py
pinned: false
license: mit

Pytorch LIghtning Model Trained for CIFAR10

• This application showcases the inference capabilities of a model trained on the CIFAR dataset.
• The architecture utilized is based on Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun's work on Deep Residual Learning for Image Recognition (arXiv:1512.03385).
• It has been recreated and trained on the CIFAR10 dataset, achieving an accuracy of 85%+ within 24 epochs. The training process was accelerated using the One Cycle Policy technique. The model implementation is done using PyTorch Lightning.
• The model is coded using PyTorch Lightning 2.2.2.

Mentioned below is the link for Training Repository Training Repo Link

• Following the training process, the model is saved locally and then uploaded to Gradio Spaces.

• Attached below is the link to download model file

• This app has two features :

• GradCam: " - To visualize the specific regions of the image that the model focuses on during inference

  • This insight can guide the development of augmentation strategies aimed at enhancing model accuracy"

• Misclassified Image:

  • Despite achieving a test accuracy of over 85% during model training, there remained a 12% misclassification rate for certain images.
  • This feature facilitates the visualization of misclassified images alongside their correct and incorrect labels.
  • Such visualization aids in devising strategies to enhance accuracy, especially for specific classes."

Usage:

GradCam

• Upload an image or choose from predefined examples.
• Adjust the opacity.
• Specify the number of top classes you wish to view.
• Click the 'Submit' button to generate the results."

Misclassified Images

• Select the number of misclassified images you want to see
• Click on Display Misclassified Images to show the images in the center and their respective correct and misclassified labels in the sequence

Thank you