Modify 'val_step()' to validate data for each val mode epoch (#123)

Co-authored-by: Nlizz <innerlee@users.noreply.github.com>

mmaction/models/recognizers/base.py

@@ -164,7 +164,7 @@ class BaseRecognizer(nn.Module, metaclass=ABCMeta):
         imgs = data_batch['imgs']
         label = data_batch['label']
 
-        losses = self.forward(imgs, label)
+        losses = self(imgs, label)
 
         loss, log_vars = self._parse_losses(losses)
 
@@ -183,9 +183,15 @@ class BaseRecognizer(nn.Module, metaclass=ABCMeta):
         not implemented with this method, but an evaluation hook.
         """
         imgs = data_batch['imgs']
+        label = data_batch['label']
+
+        losses = self(imgs, label)
 
-        results = self.forward(imgs, None, return_loss=False)
+        loss, log_vars = self._parse_losses(losses)
 
-        outputs = dict(results=results)
+        outputs = dict(
+            loss=loss,
+            log_vars=log_vars,
+            num_samples=len(next(iter(data_batch.values()))))
 
         return outputs

tools/train.py

@@ -3,6 +3,7 @@ import copy
 import os
 import os.path as osp
 import time
+import warnings
 
 import mmcv
 import torch
@@ -124,6 +125,10 @@ def main():
 
     datasets = [build_dataset(cfg.data.train)]
     if len(cfg.workflow) == 2:
+        if args.validate:
+            warnings.warn('val workflow is duplicated with `--validate`, '
+                          'it is recommended to use `--validate`. see '
+                          'https://github.com/open-mmlab/mmaction2/pull/123')
         val_dataset = copy.deepcopy(cfg.data.val)
         datasets.append(build_dataset(val_dataset))
     if cfg.checkpoint_config is not None: