README.md

metadata

license: mit
title: app gradio
sdk: gradio

model is trained using pytorch-lightning. and gives above 90 percent accuracy. training code can be found at this repo - https://github.com/srikanthp1/S12.git

bringing up the app takes time as it has to download test_downloader

we have three tabs one for each option

1) takes image as input. gets the top k and also gives the grad cam on desired layer

give an example image or other image along with transparency(opacity), layer for gradcam, top classes number

UI FOR IMAGE INPUT

(https://huggingface.co/spaces/srikanthp07/s12/tree/main/examples/cat.png)

2) misclassified images - gets misclassified images from test_loader

give number of images to display, from test_dataloader

UI FOR MISCLASSIFIED IMAGE

OUTPUT SAMPLE

3) get all images from test_loader and give grad cam

give number of images, transparency for overlap, layer number which is upto 3 3 being very spotty

GRAD-CAM UI

GRAD-CAM OUTPUT SAMPLE