This is the implementation of STG-GradCAM paper to visualize the importance of Joint-time importance map for STGCNs trained on Skeleton based activity recognition dataset
Paper: Gradient-Weighted Class Activation Mapping for Spatio Temporal Graph Convolutional Network
Author: Pratyusha Das, Antonio Ortega
Link: https://ieeexplore.ieee.org/document/9746621
Please follow the steps on STGCN_README to train and test the STGCN model Once you have the pre-procesed skeleton data and pretrained model,
cd stg-gradcam
run python main.py recognition -c config/st_gcn/ntu-xsub/test.yaml
to generate the joint time importance map for any avtivity datapoint
Once you have the joint time importance map, you can use the code in 'plot_joint_time_importance_skeleton' folder to generate the plots open matlab
cd plot_joint_time_importance_skeleton
run layerwise_gradcam_plot.m