nandeesh-n/CIFAR10_Custom_ResNet

Application Design

• —The gradio components are meticulously organised and placed as per the requirement.
• —The GradCAM and misclassification settings are segregated into different sections.
• —GradCAM settings includes the number of GradCAM images to visualize, layer of the model, and the opacity of a GradCAM image. Similarly, misclassification settings includes how many misclassified images needs to be shown.
• —Based on whether user wants to visualize the GradCAM/ misclassified images, the respective components are enabled/disabled.
• —The Input Image component accepts single image as an input and passes on to the custom ResNet model to get the predictions. The output predictions are presented in Output Label and the respective GradCAM image is displayed in the GradCAM component.
• —# of top class allows user to visualize the number of top class probablities.
• —A set of 10 example images are provided which are a part of CIFAR10 dataset.

How app is running?

• —When user uploads an image, it is fed to custom ResNet model to obtain predictions and class probabilities.
• —The GradCAM image is also generated and displayed.
• —Based on GradCAM and misclassification settings, the additional images are pulled from examples list and displayed/ visualized.