wjm-wjm

README.md

@@ -1,8 +1,6 @@
 # Optimal Transport for Label-Efficient Visible-Infrared Person Re-Identification (OTLA-ReID)
 This is Official Repository for "Optimal Transport for Label-Efficient
-Visible-Infrared Person Re-Identification", ECCV 2022.
-
-This work is done at the DMCV Laboratory of East China Normal University. You can link at [DMCV-Lab](https://dmcv-ecnu.github.io/) to find DMCV Laboratory website page.
+Visible-Infrared Person Re-Identification", ECCV 2022. This work is done at the DMCV Laboratory of East China Normal University. You can link at [DMCV-Lab](https://dmcv-ecnu.github.io/) to find DMCV Laboratory website page.
 
 ### Update:
 **[2022-7-17]** Semi-supervised setting and supervised setting can be run with current code. Unsupervised setting will be updated with a few of days.

main_train.py

@@ -26,10 +26,10 @@ def main_worker(args, args_main):
     end_epoch = args.end_epoch
 
     ## set gpu id and seed id
+    set_seed(args.seed)
     os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
     torch.backends.cudnn.benchmark = True  # accelerate the running speed of convolution network
     device = "cuda" if torch.cuda.is_available() else "cpu"
-    set_seed(args.seed)
 
     ## set file
     if not os.path.isdir(args.dataset + "_" + args.setting + "_" + args.file_name):

utils.py

@@ -287,6 +287,8 @@ def close(self):
 
 
 def set_seed(seed, cuda=True):
+    random.seed(seed)
+    np.random.RandomState(seed)
     np.random.seed(seed)
     torch.manual_seed(seed)
     if cuda: