xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-11 01:27:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix Video IO error for short video (#17791)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py 2025-05-07 23:36:06 +08:00 committed by GitHub
parent 1a6af1453d
commit be8ff88e66
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 21 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
tests/multimodal/test_utils.py
View File

@ -26,6 +26,11 @@ TEST_IMAGE_URLS = [
"https://upload.wikimedia.org/wikipedia/commons/0/0b/RGBA_comp.png", "https://upload.wikimedia.org/wikipedia/commons/0/0b/RGBA_comp.png",
] ]
TEST_VIDEO_URLS = [
"https://www.bogotobogo.com/python/OpenCV_Python/images/mean_shift_tracking/slow_traffic_small.mp4",
"https://filesamples.com/samples/video/avi/sample_640x360.avi",
]
@pytest.fixture(scope="module") @pytest.fixture(scope="module")
def url_images() -> dict[str, Image.Image]: def url_images() -> dict[str, Image.Image]:
@ -134,6 +139,18 @@ async def test_fetch_image_local_files(image_url: str):
f"file://{temp_dir}/../{os.path.basename(image_url)}") f"file://{temp_dir}/../{os.path.basename(image_url)}")
@pytest.mark.asyncio
@pytest.mark.parametrize("video_url", TEST_VIDEO_URLS)
@pytest.mark.parametrize("num_frames", [-1, 32, 1800])
async def test_fetch_video_http(video_url: str, num_frames: int):
connector = MediaConnector()
video_sync = connector.fetch_video(video_url, num_frames=num_frames)
video_async = await connector.fetch_video_async(video_url,
num_frames=num_frames)
assert np.array_equal(video_sync, video_async)
# Used for the next two tests related to `merge_and_sort_multimodal_metadata`. # Used for the next two tests related to `merge_and_sort_multimodal_metadata`.
class TestCase(NamedTuple): class TestCase(NamedTuple):
mm_positions: "MultiModalPlaceholderDict" mm_positions: "MultiModalPlaceholderDict"

6
vllm/multimodal/video.py
View File

@ -81,7 +81,8 @@ class OpenCVVideoBackend(VideoLoader):
total_frames_num = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) total_frames_num = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
full_read = num_frames == -1 or total_frames_num < num_frames full_read = num_frames == -1 or total_frames_num < num_frames
if full_read: if full_read:
frame_idx = list(range(0, total_frames_num)) num_frames = total_frames_num
frame_idx = list(range(0, num_frames))
else: else:
uniform_sampled_frames = np.linspace(0, uniform_sampled_frames = np.linspace(0,
total_frames_num - 1, total_frames_num - 1,
@ -104,7 +105,8 @@ class OpenCVVideoBackend(VideoLoader):
frames[i] = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) frames[i] = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
i += 1 i += 1
# we expect all frames loaded # we expect all frames loaded
assert i == num_frames assert i == num_frames, (f"Expected reading {num_frames} frames, "
f"but only loaded {i} frames from video.")
return frames return frames