Skip to content

Latest commit

 

History

History
128 lines (100 loc) · 6.81 KB

README.md

File metadata and controls

128 lines (100 loc) · 6.81 KB

PWC PWC

Semi-Supervised Temporal Action Detection with Proposal-Free Masking

Sauradip Nag1,2,+Xiatian Zhu1,3Yi-Zhe Song1,2Tao Xiang1,2
1CVSSP, University of Surrey, UK  2iFlyTek-Surrey Joint Research Center on Artificial Intelligence, UK 
3Surrey Institute for People-Centred Artificial Intelligence, UK
+corresponding author

Accepted to ECCV 2022

Updates

  • (June, 2022) We released SPOT training and inference code for ActivityNetv1.3 dataset.
  • (June, 2022) SPOT is accepted by ECCV 2022.

Summary

  • First single-stage proposal-free framework for Semi-Supervised Temporal Action Detection (SS-TAD) task.
  • Being single-stage, it does not suffers from the notorius Proposal Error Propagation problem.
  • Proposed a novel pre-text task for Action Detection based on the notion of Random Foreground.
  • A novel Boundary Refinement strategy is proposed based on contrastive learning.
  • With just 10% labeled videos majority of the existing TAD approaches are surpassed in terms of performance.

Abstract

Existing temporal action detection (TAD) methods rely on a large number of training data with segment-level annotations. Collecting and annotating such a training set is thus highly expensive and unscalable. Semi-supervised TAD (SS-TAD) alleviates this problem by leveraging unlabeled videos freely available at scale. However, SS-TAD is also a much more challenging problem than supervised TAD, and consequently much under-studied. Prior SS-TAD methods directly combine an existing proposal-based TAD method and a SSL method. Due to their sequential localization (e.g, proposal generation) and classification design, they are prone to proposal error propagation. To overcome this limitation, in this work we propose a novel Semi-supervised Temporal action detection model based on PropOsal-free Temporal mask (SPOT) with a parallel localization (mask generation) and classification architecture. Such a novel design effectively eliminates the dependence between localization and classification by cutting off the route for error propagation in-between. We further introduce an interaction mechanism between classification and localization for prediction refinement, and a new pretext task for self-supervised model pre-training. Extensive experiments on two standard benchmarks show that our SPOT outperforms state-of-the-art alternatives, often by a large margin.

Architecture

Getting Started

Requirements

  • Python 3.7
  • PyTorch == 1.9.0 (Please make sure your pytorch version is atleast 1.8)
  • NVIDIA GPU
  • Kornia

Environment Setup

It is suggested to create a Conda environment and install the following requirements

pip3 install -r requirements.txt

Download Features

Download the video features and update the Video paths/output paths in config/anet.yaml file. For now ActivityNetv1.3 dataset config is available. We are planning to release the code for THUMOS14 dataset soon.

Dataset Feature Backbone Pre-Training Link
ActivityNet TSN Kinetics-400 Google Drive
THUMOS TSN Kinetics-400 Google Drive
ActivityNet I3D Kinetics-400 Google Drive
THUMOS I3D Kinetics-400 Google Drive

Model Training

To train SPOT from scratch run the following command. The training configurations can be adjusted from config/anet.yaml file. This training includes both Pre-training and the fine-tuning stages.

python spot_train.py

Model Inference

We provide the pretrained models containing the checkpoint for I3D features on ActivityNetv1.3 . It can be found in the Link

After downloading the checkpoints, the checkpoints path can be saved in config/anet.yaml file. The model inference can be then performed using the following command

python spot_inference.py

Model Evaluation

To evaluate our TAGS model run the following command.

python eval.py

Performance

Qualitative Results

TO-DO Checklist

  • Support for THUMOS14 dataset
  • Enable multi-gpu training

Acknowlegement

This code repository has borrowed some parts of SSTAP and BMN. We thank the author for open-sourcing their codes and clarifying the doubts.

Citation

If you find this project useful for your research, please use the following BibTeX entry.

@article{nag2022temporal,
  title={Temporal Action Detection with Global Segmentation Mask Learning},
  author={Nag, Sauradip and Zhu, Xiatian and Song, Yi-Zhe and Xiang, Tao},
  journal={arXiv preprint arXiv:2207.06580},
  year={2022}
}
}