Does GradCAM in mmaction2 support transformer-based models?

[yehuixie]

When I use tools/visualizations/vis_cam.py for cnn-based models I can get the expected results.
However, it seems that it doesn't work for transformer-based models.
What do I need to do if I want to use Gradcam for Transformer-based models?

[cir7]

Currently, it only supports cnn-based models due to the incompatible feature shape with transformed-based models. You can try to adapt the GradCAM to transformer-based models, mainly about _calculate_localization_map and _alpha_blending.

[yehuixie]

Thanks a lot. I've solved it.