BDRC-Net

This repository is the official implementation of BDRC-Net: Boundary Discretization and Reliable Classification Network for Temporal Action Detection.

Requirements

To install requirements:

pip install -r requirements.txt

Data Preparation

Download Thumos14 from BaiDuYun (code: bdrc).

Please ensure the data structure is as below:

├── data
   └── thumos
       ├── val
           ├── video_validation_0000051_02432.npz
           ├── video_validation_0000051_02560.npz
           └── ...
       └── test
           ├── video_test_0000004_00000.npz
           ├── video_test_0000004_00256.npz
           └── ...
       └── detclasslist.txt
       └── th14_groundtruth.json

Training

To train the BDRC-Net model on THUMOS'14 dataset, please first modify parameters in:

./experiments/BDRCNet_thumos.yaml

Then run this command:

python main.py --cfg experiments/BDRCNet_thumos.yaml

Pre-trained Models

You can download pretrained models here:

• THUMOS14, (code: bdrc), trained on THUMOS'14 using parameters same as "./experiments/BDRCNet_thumos.yaml".

Test

The command for test is

python model_inference.py --cfg experiments/BDRCNet_thumos.yaml --resume output/thumos_cc/model_34.pth

Acknowledgement

Our code is based on A2Net. We would like to express our gratitude for this outstanding work.