These are official codes for ICASSP2022 paper "LOOK, LISTEN AND PAY MORE ATTENTION: FUSING MULTI-MODAL INFORMATION FOR VIDEO VIOLENCE DETECTION"

Overall Structure

[September 30th] Our codes and more details about the work will be released soon.

XD-Violence XD-Violence This dataset is from the official GitHub of "Not only Look, but also Listen: Learning Multimodal Violence Detection under Weak Supervision (ECCV2020)".

Based on their ground truth, we have created a new format frame_mask of the test dataset for our method and published it in the repository.

Visualization results of violence scenes

Visualization results of fused attention

• python>=3.6
• pytorch=1.5.0+cu101
• torchvision=0.6.0+cu101
• scikit-learn
• numpy
• apex
• opencv