Learning to Detect Violent Videos using Convolutional Long Short-Term Memory

Pytorch version can be found here

The source code associated with the paper Learning to Detect Violent Videos using Convolutional Long Short-Term Memory, published in AVSS-2017.

The code for evaluating the performance of the model proposed in the paper is given. The convLSTM model is based on the source provided by the authors of the paper Spatio-temporal video autoencoder with differentiable memory

Prerequisites

• Torch7(http://torch.ch/docs/getting-started.html)

How to run

The pre-trained models can be downloaded from the following links:

• Hockey dataset
• Movies dataset
• Violentflows dataset

To evaluate the performance of the model trained run the following command, th main-run.lua -FightsDataset path-to-fights-video-file -noFightsDataset path-to-nonfights-video-file -model path-to-model -nSeq number-of-frames

The dataset should be in the format number_of_videosxnumber_of_framesx3xwidthxheight

To cite our paper/code:

@inproceedings{sudhakaran2017learning,
  title={Learning to detect violent videos using convolutional long short-term memory},
  author={Sudhakaran, Swathikiran and Lanz, Oswald},
  booktitle={Advanced Video and Signal Based Surveillance (AVSS), 2017 14th IEEE International Conference on},
  pages={1--6},
  year={2017},
  organization={IEEE}
}