search

layumi / Person_reID_baseline_pytorch

:bouncing_ball_person: Pytorch ReID: A tiny, friendly, strong pytorch implement of person re-id / vehicle re-id baseline. Tutorial 👉https://github.com/layumi/Person_reID_baseline_pytorch/tree/master/tutorial
https://www.zdzheng.xyz
MIT License
4.13k stars 1k forks source link

A question about Occluded-Duke dataset #345

Open ZzzybEric opened 1 year ago

ZzzybEric commented 1 year ago

hello , layumi I want to train the occluded duke dataset , can i use prepare.py to prepare dataset directly?

And i use prepare.py to prepare occluded-duke dataset , use swin transformer backbone , rank1 achieved 59%, do you have any suggestions to improve this result? thank you!!

layumi commented 1 year ago

Hi @ZzzybEric

Thank you.

  1. I am not sure about this. You need to check the image number of train / query/ gallery with the paper to ensure it is correct.

  2. You may consider to use more losses / PCB / random erasing / reranking to improve result. Of course, changing lr Rate and batch size are always the first step.