This is a pytorch implementation of TURN TAP: Temporal Unit Regression Network for Temporal Action Proposals. This code is for research purpose and suggestions are welcome.
Pytorch 0.4.0
CUDA 8.0
Python 2.7.6
The tensorflow implementation code provided by the authors: https://github.com/jiyanggao/TURN-TAP/tree/master/turn_codes
DenseFlow Features in the Google Drive:val set, test set
use git to clone this repository
$ git clone --recursive https://github.com/JunxuanZhang/TURN-TAP-pytorch/
Then create two necessary folders
$ mkdir features results
Move the downloaded features to the 'features' folder
To train and evaluate the TURN model, run the 'main.py' script
$ python main.py
if you want to continue training from the specfic checkpoint, use '--resume' option
$ python main.py --resume CHECKPOINT_PATH