cuhk03's protocol about alignedreid-re-production-pytorch

[huanghoujing]

In fact, there is no standard train/test split for CUHK03 in recent papers.

1. If you want to use the new train/test protocol as I provide here (which is popular in recent papers because it provides much larger test set), you only have to set the dataset argument to cuhk03 when running the script script/tri_loss/train.py.

2. If you want to use the train/test protocol in the paper. At the end of Section 4.1 of the AlignedReID paper, it says

   We only randomly split the dataset once for training and testing, and the gallery includes 200 identities.

   So when transforming CUHK03 using script/dataset/transform_cuhk03.py, you only have to provide your own train_test_partition_file corresponding to this split. This train_test_partition_file specifies image names for train, query and gallery sets.

[liangbh6]

@huanghoujing Nice explanation! Thanks.