Awesome

TranS4mer

This is the PyTorch Implementation of Efficient Movie Scene Detection using State-Space Transformers (TranS4mer) [arxiv]

1. Data and Environmental Setup

We have tested the implementation on the following environment:

• Python 3.8.12 / PyTorch 1.10.0 / torchvision 0.11.1 / CUDA 11.3

The code is based on BaSSL Follow BaSSL for environmental setup and data download.

Also follow S4 for installation regarding S4 models.

2. Training

(1) Pre-training BaSSL
cd trans4mer; bash ../scripts/run_pretrain_bassl.sh

(2) Finetuning and Evaluation

cd trans4mer; bash ../scripts/run_finetune.sh

3. Pre-trained Models

TODO

4. Citation

If you find this code helpful for your research, please cite our paper.

@inproceedings{islam2023efficient,
  title={Efficient Movie Scene Detection using State-Space Transformers},
  author={Islam, Md Mohaiminul and Hasan, Mahmudul and Athrey, Kishan Shamsundar and Braskich, Tony and Bertasius, Gedas},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={18749--18758},
  year={2023}
}