Visual stream features vs Spatial-Temporal stream

Wanggcong / Spatial-Temporal-Re-identification

[AAAI 2019] Spatial Temporal Re-identification

MIT License

384 stars 77 forks source link

Visual stream features vs Spatial-Temporal stream #28

Closed saif86 closed 4 years ago

saif86 commented 4 years ago

Hello. Pretty nice and well coded repo. Thanks for your contributions. I have a question regarding feature extractions. For visual stream you are extracting gallery and query features after training the model. But for Spatial-Temporal stream, you are extracting distribution from training set data(camID+frames+labels). Shouldn't it be from gallery/query set? Because later you are joining those features. Please explain I have a little confusion. Many thanks

Wanggcong commented 4 years ago

--For Spatial-Temporal stream, you are extracting distribution from training set data(camID+frames+labels)? Yes! Shouldn't it be from gallery/query set? --No. We assume the examples of training/gallery/query sets are drawn from the same distributions.

saif86 commented 4 years ago

OK! Thanks a lot for your quick reply.