Use Grad-Cam on Pytorch Project
Budget: $30 – $250 USD
I am looking for a freelancer who has intermediate experience with PyTorch and Grad-Cam. I would like to visualize the activation maps of a sample from the Dev set to understand the model predictions. Particularly, I would like to visualize the saliency maps of the conv2d module (ResNet18) used in this project on Github https://github.com/ycmin95/VAC_CSLR/tree/main
The project authors uploaded the trained model here :
https://www.dropbox.com/s/zbas78emfz5m4bp/resnet18_slr_pretrained_distill25.pt?dl=0
I have uploaded a couple of samples from the Dev set to test your code here:
https://drive.google.com/file/d/1agsLRjKACbkQK1UdeiGGfQqdS1cKd4zl/view?usp=sharing
Link to the Gard-cam library: https://github.com/jacobgil/pytorch-grad-cam
The desired output should be like Figure 4 (middle) in this paper:
https://openaccess.thecvf.com/content/CVPR2023/papers/Zheng_CVT-SLR_Contrastive_Visual-Textual_Transformation_for_Sign_Language_Recognition_With_Variational_CVPR_2023_paper.pdf
The implemented code should have the following features:
1- Enable the user to specify the path to the input folder in which the Grad-cam would be applied.
2- The user should be able to select the Grad-cam method, with available method choices (if applicable): GradCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, EigenGradCAM, LayerCAM, FullGrad.
3- The output images should be saved in the same directory.
This usage example from the Grad-Cam project maybe helpful:
https://github.com/jacobgil/pytorch-grad-cam/blob/master/usage_examples/swinT_example.py
The project authors uploaded the trained model here :
https://www.dropbox.com/s/zbas78emfz5m4bp/resnet18_slr_pretrained_distill25.pt?dl=0
I have uploaded a couple of samples from the Dev set to test your code here:
https://drive.google.com/file/d/1agsLRjKACbkQK1UdeiGGfQqdS1cKd4zl/view?usp=sharing
Link to the Gard-cam library: https://github.com/jacobgil/pytorch-grad-cam
The desired output should be like Figure 4 (middle) in this paper:
https://openaccess.thecvf.com/content/CVPR2023/papers/Zheng_CVT-SLR_Contrastive_Visual-Textual_Transformation_for_Sign_Language_Recognition_With_Variational_CVPR_2023_paper.pdf
The implemented code should have the following features:
1- Enable the user to specify the path to the input folder in which the Grad-cam would be applied.
2- The user should be able to select the Grad-cam method, with available method choices (if applicable): GradCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, EigenGradCAM, LayerCAM, FullGrad.
3- The output images should be saved in the same directory.
This usage example from the Grad-Cam project maybe helpful:
https://github.com/jacobgil/pytorch-grad-cam/blob/master/usage_examples/swinT_example.py