resolve issues

Author: kprokofi

configs/classification/efficientnet_v2_b0.yml

@@ -82,7 +82,7 @@ train:
     enable: True
     ema_decay: 0.999
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

configs/multilabel_classification/efficientnet_v1_b0.yml

@@ -19,14 +19,14 @@ model:
     p: 0.1
 
 custom_datasets:
-  roots: ['/ssd/datasets/nus_wide/train.json', '/ssd/datasets/nus_wide/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['nus_wide_train', 'nus_wide_val']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0.485, 0.456, 0.406]

configs/multilabel_classification/efficientnet_v2_b0.yml

@@ -19,14 +19,14 @@ model:
     p: 0.1
 
 custom_datasets:
-  roots: ['/ssd/datasets/nus_wide/train.json', '/ssd/datasets/nus_wide/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['nus_wide_train', 'nus_wide_val']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0.485, 0.456, 0.406]
@@ -80,7 +80,7 @@ train:
     enable: True
     ema_decay: 0.9995
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

configs/multilabel_classification/efficientnet_v2_small.yml

@@ -19,14 +19,14 @@ model:
     p: 0.1
 
 custom_datasets:
-  roots: ['/ssd/datasets/nus_wide/train.json', '/ssd/datasets/nus_wide/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['nus_wide_train', 'nus_wide_val']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0.5, 0.5, 0.5]
@@ -80,7 +80,7 @@ train:
     enable: True
     ema_decay: 0.9995
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

configs/multilabel_classification/mobilenetv3_large.yml

@@ -19,14 +19,14 @@ model:
     p: 0.1
 
 custom_datasets:
-  roots: ['/data/train.json', '/data/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['train_data', 'val_data']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0, 0, 0]
@@ -80,7 +80,7 @@ train:
     enable: True
     ema_decay: 0.9997
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

configs/multilabel_classification/mobilenetv3_large_75.yml

@@ -19,14 +19,14 @@ model:
     p: 0.1
 
 custom_datasets:
-  roots: ['/data/train.json', '/data/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['train_data', 'val_data']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0.485, 0.456, 0.406]
@@ -80,7 +80,7 @@ train:
     enable: True
     ema_decay: 0.9995
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

configs/multilabel_classification/mobilenetv3_small.yml

@@ -17,14 +17,14 @@ model:
   save_all_chkpts: False
 
 custom_datasets:
-  roots: ['/data/train.json', '/data/val.json']
+  roots: ['data/coco/train.json', 'data/coco/val.json']
   types: ['multilabel_classification', 'multilabel_classification']
-  names: ['train_data', 'val_data']
+  names: ['coco_train', 'coco_val']
 
 data:
   root: './'
-  sources: ['train_data']
-  targets: ['val_data']
+  sources: ['coco_data']
+  targets: ['coco_data']
   height: 448
   width: 448
   norm_mean: [0.485, 0.456, 0.406]
@@ -78,7 +78,7 @@ train:
     enable: True
     ema_decay: 0.9995
   mix_precision: True
-  
+
 test:
   batch_size: 128
   evaluate: False

torchreid/utils/torchtools.py

@@ -21,26 +21,26 @@
     'load_pretrained_weights', 'ModelEmaV2'
 ]
 
+
+def params_to_device(param, device):
+    def tensor_to_device(param, device):
+        param.data = param.data.to(device)
+        if param._grad is not None:
+            param._grad.data = param._grad.data.to(device)
+
+    if isinstance(param, torch.Tensor):
+        tensor_to_device(param, device)
+    elif isinstance(param, dict):
+        for subparam in param.values():
+            tensor_to_device(subparam, device)
+
 def optimizer_to(optim, device):
     for param in optim.state.values():
-        # Not sure there are any global tensors in the state dict
-        if isinstance(param, torch.Tensor):
-            param.data = param.data.to(device)
-            if param._grad is not None:
-                param._grad.data = param._grad.data.to(device)
-        elif isinstance(param, dict):
-            for subparam in param.values():
-                if isinstance(subparam, torch.Tensor):
-                    subparam.data = subparam.data.to(device)
-                    if subparam._grad is not None:
-                        subparam._grad.data = subparam._grad.data.to(device)
+        params_to_device(param, device)
 
 def scheduler_to(sched, device):
     for param in sched.__dict__.values():
-        if isinstance(param, torch.Tensor):
-            param.data = param.data.to(device)
-            if param._grad is not None:
-                param._grad.data = param._grad.data.to(device)
+        params_to_device(param, device)
 
 def save_checkpoint(
     state, save_dir, is_best=False, remove_module_from_keys=False, name='model'