CTRNet

The code of "Content Temporal Relation Network for Temporal Action Proposal Generation"

Contents

---

• Usage Guide
  • Prerequisites
  • Code and Data Preparation
    • Get the code
    • Download Features
  • Training CTRNet
  • Testing CTRNet

Usage Guide

Prerequisites

The training and testing in CTRNet is reimplemented in PyTorch for the ease of use.

• [PyTorch 1.8.0][pytorch]
• [Fastai 1.0.61][fastai]

Code and Data Preparation

Get the code

Clone this repo with git, please remember to use --recursive

git clone --recursive https://github.com/YanZhang-bit/CTRNet

Download Features

THUMOS14 feature can be download here.

ActivityNet-1.3 feature can be download here. And people need rescaled the feature length of all videos to same length 100.

Training CTRNet

• To train our CTRNet on the THUMOS14 dataset

cd Thumos
python fastai_train.py

• To train our CTRNet on the ActivityNet-1.3 dataset

cd ActivityNet
python fastai_train_content.py

Testing CTRNet

• To test our CTRNet on the THUMOS14 dataset

  We provide the final model for THUMOS14 dataset Baidu Cloud (password: eyy6)

cd Thumos
python fastai_inference.py --checkpoint best_model/final_model.pth
python postprocess.py
python eval.py

• To test our CTRNet on the ActivityNet-1.3 dataset

  We provide the final model for ActivityNet-1.3 dataset Baidu Cloud (password: 3qtc)

cd ActivityNet
python inference_content.py --mode inference --checkpoint best_model/final_model.pth