Generate the training/test set of CUHK03 according to AlignedReid paper about alignedreid-re-production-pytorch

[huanghoujing]

You may ask the author to provide you the training, query and gallery image split. Then you can map the split to the format in this project.

[huanghoujing]

Thanks for your attention.
For your need, when transforming CUHK03 using script/dataset/transform_cuhk03.py, you can provide your own train_test_partition_file corresponding to the paper's split. This train_test_partition_file specifies image names for train, query and gallery sets. In fact, I am not sure about the detail split of the paper and sorry for that.

[ZHHJemotion]

@huanghoujing Thanks a lot for your reply. I would ask the author for some details about the split. Thanks again.

[ZHHJemotion]

@huanghoujing I have generated the training/test set based on the split strategy in AlignedReID paper by myself: 1267 ids for training and 200 ids for testing. Finally, I got the rank-1 acc after re-ranking with mutual learning of 94.22% and without mutual learning of 92.41%, they are a litter worse than in paper.

[huanghoujing]

Well, I am impressed by the 92%+ performance! You provide some reference for later similar experiments. Thanks a lot.

[ZHHJemotion]

Thank you for your reply. How can I provide these reference?

[huanghoujing]

Oh, I may not express it well. I was not asking you to provide anything. I just meant that your comments here may be useful for people doing the same experiments. Forgive me for my expression.

[ZHHJemotion]

I forked it and added the code of new split training/testing set, so that someone can reference. Thank you!