search

dingfengshi / TriDet

[CVPR2023] Code for the paper, TriDet: Temporal Action Detection with Relative Boundary Modeling
MIT License
160 stars 13 forks source link

Data acquisition problem #4

Closed YCA-eng closed 1 year ago

YCA-eng commented 1 year ago

Thanks for your excellent work. How to download i3d_feature and slowfast_feature in the same format as yours.

dingfengshi commented 1 year ago

Hi! For the slowfast_feature, you need to download all the files end with (.validation.tar.bz2. and .training.tar.bz2.) in the folder hacs-data/features from their BaiduYun link. Then you can merge all the volumes for training or validation dataset and unzip the merged file, for example 

# merge volumes
cat slowfast101.epoch9.87.52.finetune.pool.t.keep.t.s8.validation.tar.bz2* > slowfast101.epoch9.87.52.finetune.pool.t.keep.t.s8.validation.tar.bz2

#unzip the new file
tar xjvf slowfast101.epoch9.87.52.finetune.pool.t.keep.t.s8.validation.tar.bz2

You can try that, sorry I have forgotten the specific command executed at that time. If there is anything wrong with this command please let me know, thank you!