riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 33 Packages Projects Releases Wiki Activity

Fix temporal padding in Qwen2VLImageProcessor when the number of frames is not divisible by temporal_patch_size (#38076)

Browse Source 
Qwen2VL: Fix temporal padding in Qwen2VLImageProcessor when frames are not divisible by temporal_patch_size
This commit is contained in:
Ritwick Chaudhry 2025-05-14 03:28:21 -07:00 committed by GitHub
parent aaf224d570
commit fe918d13b9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 21 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/transformers/models/qwen2_vl/image_processing_qwen2_vl.py
View File

@ -274,7 +274,9 @@ class Qwen2VLImageProcessor(BaseImageProcessor):
if data_format == ChannelDimension.LAST:
patches = patches.transpose(0, 3, 1, 2)
if patches.shape[0] % temporal_patch_size != 0:
repeats = np.repeat(patches[-1][np.newaxis], temporal_patch_size - 1, axis=0)
repeats = np.repeat(
patches[-1][np.newaxis], temporal_patch_size - (patches.shape[0] % temporal_patch_size), axis=0
)
patches = np.concatenate([patches, repeats], axis=0)
channel = patches.shape[1]
grid_t = patches.shape[0] // temporal_patch_size

18
tests/models/qwen2_vl/test_image_processing_qwen2_vl.py
View File

@ -312,6 +312,24 @@ class Qwen2VLImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
expected_output_video_shape = [112, 1176]
self.assertListEqual(list(prcocess_out.pixel_values.shape), expected_output_video_shape)
def test_temporal_padding(self):
for image_processing_class in self.image_processor_list:
# Initialize image_processing
image_processing = image_processing_class(**self.image_processor_dict)
# Create random video inputs with a number of frames not divisible by temporal_patch_size
image_processor_tester = Qwen2VLImageProcessingTester(self, num_frames=5, temporal_patch_size=4)
video_inputs = image_processor_tester.prepare_video_inputs(equal_resolution=True)
# Process the video inputs
process_out = image_processing(None, videos=video_inputs, return_tensors="pt")
encoded_video = process_out.pixel_values_videos
# Check the shape after padding
expected_output_video_shape = (102900, 1176) # Adjusted based on padding
self.assertEqual(tuple(encoded_video.shape), expected_output_video_shape)
# Check divisibility by temporal_patch_size
self.assertEqual(encoded_video.shape[0] % 4, 0)
@require_vision
@require_torch
def test_slow_fast_equivalence(self):