SlowFast Network for Continuous Sign Language Recognition

This repository provides the official pytorch implementation of our paper:

SlowFast Network for Continuous Sign Language Recognition

Junseok Ahn, Youngjoon Jang, Joon Son Chung

Accepted to ICASSP 2024

This repo is based on VAC (ICCV 2021), CorrNet (CVPR 2023), and PySlowFast.

Abstract

Fig. 1. Overall Architecture. Our framework extracts spatial and dynamic features in parallel using the Slow and Fast pathways. Bi-directional Feature Fusion (BFF) facilitates the exchange of rich information between two pathways. Pathway Feature Enhancement (PFE) amplifies the sequence feature of each pathway by utilising characteristics sourced from different sample rates.

The objective of this work is the effective extraction of spatial and dynamic features for Continuous Sign Language Recognition (CSLR). To accomplish this, we utilise a two-pathway SlowFast network, where each pathway operates at distinct temporal resolutions to separately capture spatial (hand shapes, facial expressions) and dynamic (movements) information. In addition, we introduce two distinct feature fusion methods, carefully designed for the characteristics of CSLR: (1) Bi-directional Feature Fusion (BFF), which facilitates the transfer of dynamic semantics into spatial semantics and vice versa; and (2) Pathway Feature Enhancement (PFE), which enriches dynamic and spatial representations through auxiliary subnetworks, while avoiding the need for extra inference time. As a result, our model further strengthens spatial and dynamic representations in parallel. We demonstrate that the proposed framework outperforms the current state-of-the-art performance on popular CSLR datasets, including PHOENIX14, PHOENIX14-T, and CSL-Daily.

Prerequisites

• Python == 3.8
• PyTorch == 1.13
• ctcdecode == 1.0.3 [parlance/ctcdecode]，for beam search decode.
• You can install other required modules by running
  pip install -r requirements.txt

Data Preparation

Please follow the instruction in CorrNet github repo to download and preprocess the datasets (PHOENIX2014, PHOENIX2014-T, CSL-Daily). The structure of dataset directory is as follows (There may be other additional directories.):

dataset
├── phoenix2014
│   └── phoenix-2014-multisigner
│       └── features
│           └── fullFrame-256x256px
├── phoenix2014-T
│   └── features
│       └── fullFrame-256x256px
└── CSL-Daily
    └── sentence
        └── frames_256x256

Inference

To evaluate the pretrained model, run the command below:

python main.py --device 0 --dataset phoenix2014 --phase test --load-weights ./best_checkpoints/phoenix2014_dev_18.01_test_18.28.pt --work-dir ./work_dir/phoenix2014_test/

Training

Before you start training, download the pre-trained SlowFast pkl file by running the following code:

mkdir ckpt && cd ckpt
wget https://dl.fbaipublicfiles.com/pyslowfast/model_zoo/ava/pretrain/SLOWFAST_64x2_R101_50_50.pkl

To train the SlowFastSign model, run the command below:

python main.py --device 0 --dataset phoenix2014 --loss-weights Slow=0.25 Fast=0.25 --work-dir ./work_dir/phoenix2014/

You can check other arguments for training from utils/parameters.py

Citation

If you find our work useful for your research, please cite our work with the following bibtex:

@inproceedings{ahn2024slowfast,
  author={Ahn, Junseok and Jang, Youngjoon and Chung, Joon Son},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing}, 
  title={Slowfast Network for Continuous Sign Language Recognition}, 
  year={2024},
  pages={3920-3924},
  doi={10.1109/ICASSP48485.2024.10445841}
}