Giters

huanghoujing / AlignedReID-Re-Production-Pytorch

Reproduce AlignedReID: Surpassing Human-Level Performance in Person Re-Identification, using Pytorch.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

What is the difference between this project and person-reid-triplet-loss-baseline?

cesarandreslopez opened this issue · comments

commented

@huanghoujing you seem to be actively maintaining both projects and both work very well. I'm wondering what are the main differences between both projects?

commented

Hi, thanks for your attention.

  • Repository AlignedReID-Re-Production-Pytorch is mainly for exploring the algorithms used in paper AlignedReID, including shortest distance and mutual learning. So it contains some customized code that is only necessary when using AlignedReID.
  • Repository person-reid-triplet-loss-baseline is a clean version of training triplet loss, which is straightforward to read.
  • The triplet loss baselines in two repositories are the same, except that person-reid-triplet-loss-baseline also tries a trick (i.e. larger resolution in the last conv layer of resnet-50) and achieves good results. So if someone is looking for a strong triplet loss baseline, I think person-reid-triplet-loss-baseline is a good one.
commented

Excellent! Thank you!