The problem of the intersection of person ids betweem trainval and gallery about person-reid-triplet-loss-baseline

[huanghoujing]

Thank you a lot for pointing out the problem. I will look into that!

[huanghoujing]

1. In your code, trainval_pids would be updated by both detected and labeled parts of the dataset, similarly for query_pids and gallery_pids. When you check the detected and labeled parts independently, as in the modified code below, you would find there is no problem.

if osp.exists(train_test_partition_file):
  train_test_partition = load_pickle(train_test_partition_file)
else:
  raise RuntimeError("Train/test partition file should be provided.'")

for im_type in ['detected', 'labeled']:
  trainval_pids = set()
  query_pids = set()
  gallery_pids = set()

  trainval_im_names = train_test_partition[im_type]['train_im_names']
  query_im_names = train_test_partition[im_type]['query_im_names']
  gallery_im_names = train_test_partition[im_type]['gallery_im_names']

  trainval_pids.update(set([parse_im_name(n, 'id') for n in trainval_im_names]))
  query_pids.update(set([parse_im_name(n, 'id') for n in query_im_names]))
  gallery_pids.update(set([parse_im_name(n, 'id') for n in gallery_im_names]))

  print(trainval_pids & gallery_pids)
  assert query_pids <= gallery_pids
  assert trainval_pids.isdisjoint(gallery_pids)

2. Your code indicates the inconsistency between the detected partitions and labeled partitions. This originates from the official new train/test protocol. I provide a script to verify this. Where you will find that

• detected train ids share 766 ids with labeled train ids
• detected test ids share 699 ids with labeled test ids
• id (4, 30) (id 1201 in my project) is in train ids for detected, but in test ids for labeled
• id (2, 359) (id 1389 in my project) is in test ids for detected, but in train ids for labeled

3. To make the generation of my re_ranking_train_test_split.pkl more clear, and to make future problems more trackable, I provide details of the generation process in this repository.