Apart from human action can we also detect the objects in the frame??
BakingBrains opened this issue · 6 comments
Thanks for playing with Activity2Vec!
Activity2Vec aims at understanding human actions in general scenarios (w/ or w/o interacted objects). Therefore, we simplified the codebase and excluded the object detection module. See our previous codebase HAKE-Action if you need it.
Thanks for the reply. There are many repo's in HAKE-Action, can you please specify any. I have gone through TIN from HAKE. I found it very interesting. Is there any better guide to setup and train or inferencing.
Can I use it for real-time action detection and also is there any way to optimize the time to process frames??
Thanks for the reply. There are many repo's in HAKE-Action, can you please specify any. I have gone through TIN from HAKE. I found it very interesting. Is there any better guide to setup and train or inferencing.
If you want to use TIN as your pipeline, you can follow the independent codebase Transferable-Interactiveness-Network.
Can I use it for real-time action detection and also is there any way to optimize the time to process frames??
We've given a demo program in Activity2Vec, and you can analyze the bottleneck of it in your scenario for optimization, since I have no knowledge about your application.
Thanks a lot @hwfan, will get back if any doubts.