Fix the audio dimensions for eval input.

PiperOrigin-RevId: 420825486

Fix the audio dimensions for eval input.
PiperOrigin-RevId: 420825486
10806935 · Liangzhe Yuan · A. Unique TensorFlower · cf1f8a2a · 10806935
Commit 10806935 authored Jan 10, 2022 by Liangzhe Yuan Committed by A. Unique TensorFlower Jan 10, 2022
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

official/vision/beta/dataloaders/video_input.py official/vision/beta/dataloaders/video_input.py +1 -1

No files found.
--- a/official/vision/beta/dataloaders/video_input.py
+++ b/official/vision/beta/dataloaders/video_input.py
@@ -361,7 +361,7 @@ class Parser(parser.Parser):
      audio = decoded_tensors[self._audio_feature]
      audio = tf.cast(audio, dtype=self._dtype)
      audio = preprocess_ops_3d.sample_sequence(
-          audio, 20, random=False, stride=1)
+          audio, self._audio_shape[0], random=False, stride=1)
      audio = tf.ensure_shape(audio, self._audio_shape)
      features['audio'] = audio