search

yjxiong / temporal-segment-networks

Code & Models for Temporal Segment Networks (TSN) in ECCV 2016
BSD 2-Clause "Simplified" License
1.54k stars 477 forks source link

How to train on untrimmed datasets？ #240

Closed VSunN closed 6 years ago

VSunN commented 6 years ago

I want to know how to train on untrimmed datasets, such as ActivityNet, Thumos14, especially how to use the temporal annotations?

wanglimin commented 6 years ago

Our UntrimmedNet is a weakly supervised framework, which dose not require temporal annotations. Please read our paper carefully.

VSunN commented 6 years ago

Thanks for your reply. I know that UntrimmedNet is a weakly supervised framework. I thought the other repository (submission to anet2016) is trained with the temporal annotations, but seems that it only use the temporal annotations to get clipped activity instances and the new aggregation is more important. Thank you very much.