search

zhangxinyu-xyz / PAST-ReID

Self-Training with Progressive Augmentation for Unsupervised Cross-Domain Person Re-identification (ICCV2019)
MIT License
99 stars 24 forks source link

How to generate prepare initial files for different data sets? #10

Open RitaDel opened 3 years ago

RitaDel commented 3 years ago

Can you tell me how to generate initial files for different data sets? Thank you

zhangxinyu-xyz commented 3 years ago

Can you tell me how to generate initial files for different data sets? Thank you

The initial model is trained from the https://github.com/syfafterzy/PCB_RPP_for_reID with PCB setting. If you want to use your own dataset, the steps are as follows:

  1. Use your own data to train the PCB model with https://github.com/syfafterzy/PCB_RPP_for_reID.
  2. Extract features of the training set from the above model.
  3. Calculate the reranked distmat as in https://github.com/zhangxinyu-xyz/PAST-ReID/blob/67f1f7a780e869aa7867167538edb03faa96dec5/reid/eval/rerankor.py#L119
Chen-Rjie commented 2 years ago

Can the author publish the PCB model of pre training? Should the number of source domain pre training hours be changed from 6 to 9? We look forward to your answer! 请问作者能否公布预训练的pcb模型？源域预训练时分支数是否应该把6修改为9呢，期待您的回答！