Paper presented at ICPR 2024 by Suvramalya Basak. The work was titled "Multi-teacher Importance Preserving Knowledge Distillation for Early Violence Prediction".
The main contributions of the paper are as follows:- We propose a lightweight two-stream violence recognition model, called EVNet, which leverages our efficient SESNet architecture.
- We train our EVNet model in a teacher-student setup to incorporate early violence prediction capability in our violence classifier. To the best of our knowledge, this is the first work to deal with early violence prediction.
- An importance preserving multi-teacher knowledge distillation is proposed to further improve violence prediction performance from partially observed videos.