Fixing frames and preprocessing.

Author: datumbox

references/video_classification/train.py

@@ -189,7 +189,7 @@ def main(args):
             split="val",
             step_between_clips=1,
             transform=transform_test,
-            frame_rate=15,
+            frame_rate=args.frame_rate,
             extensions=(
                 "avi",
                 "mp4",

torchvision/models/video/mvit.py

@@ -434,7 +434,7 @@ def _mvit(
 class MViT_V1_B_Weights(WeightsEnum):
     KINETICS400_V1 = Weights(
         url="https://download.pytorch.org/models/mvit_v1_b-a3d8bcb8.pth",
-        transforms=partial(VideoClassification, crop_size=(224, 224), resize_size=(224, 224)),
+        transforms=partial(VideoClassification, crop_size=(224, 224), resize_size=(256,)),
         meta={
             "min_size": (224, 224),
             "min_temporal_size": 16,