Spaces:
Running
on
Zero
Running
on
Zero
updated fps
Browse files- __pycache__/body_features.cpython-38.pyc +0 -0
- __pycache__/crop_face.cpython-38.pyc +0 -0
- __pycache__/crop_hands.cpython-38.pyc +0 -0
- __pycache__/dinov2_features.cpython-38.pyc +0 -0
- __pycache__/inference.cpython-38.pyc +0 -0
- __pycache__/kpe_mediapipe.cpython-38.pyc +0 -0
- __pycache__/shubert.cpython-38.pyc +0 -0
- app.py +2 -1
- features.py +4 -3
__pycache__/body_features.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/body_features.cpython-38.pyc and b/__pycache__/body_features.cpython-38.pyc differ
|
|
|
__pycache__/crop_face.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/crop_face.cpython-38.pyc and b/__pycache__/crop_face.cpython-38.pyc differ
|
|
|
__pycache__/crop_hands.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/crop_hands.cpython-38.pyc and b/__pycache__/crop_hands.cpython-38.pyc differ
|
|
|
__pycache__/dinov2_features.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/dinov2_features.cpython-38.pyc and b/__pycache__/dinov2_features.cpython-38.pyc differ
|
|
|
__pycache__/inference.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/inference.cpython-38.pyc and b/__pycache__/inference.cpython-38.pyc differ
|
|
|
__pycache__/kpe_mediapipe.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/kpe_mediapipe.cpython-38.pyc and b/__pycache__/kpe_mediapipe.cpython-38.pyc differ
|
|
|
__pycache__/shubert.cpython-38.pyc
CHANGED
|
Binary files a/__pycache__/shubert.cpython-38.pyc and b/__pycache__/shubert.cpython-38.pyc differ
|
|
|
app.py
CHANGED
|
@@ -502,10 +502,11 @@ This app uses TTIC's foundation model SHuBERT (introduced in an ACL 2025 paper,
|
|
| 502 |
|
| 503 |
**Requirements:**
|
| 504 |
- We recommend that videos be under 20 seconds. Performance for longer videos has not been tested.
|
| 505 |
-
- The signer should be the main part of the video. Videos recorded from a phone camera, tablet, or personal computer should work well. Studio recordings where the signer is farther from the camera may not work as well.
|
| 506 |
- Supported formats: MP4, MOV
|
| 507 |
|
| 508 |
**Note:**
|
|
|
|
| 509 |
- Videos will be deleted after the output is generated.
|
| 510 |
- Inquires or Feedback? Please email us at [email protected]
|
| 511 |
"""
|
|
|
|
| 502 |
|
| 503 |
**Requirements:**
|
| 504 |
- We recommend that videos be under 20 seconds. Performance for longer videos has not been tested.
|
| 505 |
+
- The signer should be the main part (like 90% space-wise) of the video. Videos recorded from a phone camera, tablet, or personal computer should work well. Studio recordings where the signer is farther from the camera may not work as well.
|
| 506 |
- Supported formats: MP4, MOV
|
| 507 |
|
| 508 |
**Note:**
|
| 509 |
+
- This is just a demo of a research project, and should NOT be used to replace an interpreter in any way.
|
| 510 |
- Videos will be deleted after the output is generated.
|
| 511 |
- Inquires or Feedback? Please email us at [email protected]
|
| 512 |
"""
|
features.py
CHANGED
|
@@ -51,8 +51,9 @@ class SHuBERTProcessor:
|
|
| 51 |
signer_video = decord.VideoReader(video_path)
|
| 52 |
|
| 53 |
signer_video_fps = signer_video.get_avg_fps()
|
| 54 |
-
target_fps = 12
|
| 55 |
-
stride = max(1, int(round(signer_video_fps / target_fps)))
|
|
|
|
| 56 |
index_list = list(range(0, len(signer_video), stride))
|
| 57 |
signer_video = signer_video.get_batch(index_list)
|
| 58 |
signer_video = signer_video.asnumpy()
|
|
@@ -106,7 +107,7 @@ if __name__ == "__main__":
|
|
| 106 |
|
| 107 |
# input_clip = "/share/data/pals/shester/datasets/openasl/clips_bbox/J-0KHhPS_m4.029676-029733.mp4"
|
| 108 |
# input_clip = "/share/data/pals/shester/inference/recordings/sabrin30fps.mp4"
|
| 109 |
-
input_clip = "/share/data/pals/shester/inference/recordings/
|
| 110 |
processor = SHuBERTProcessor(config)
|
| 111 |
output_text = processor.process_video(input_clip)
|
| 112 |
print(f"The English translation is: {output_text}")
|
|
|
|
| 51 |
signer_video = decord.VideoReader(video_path)
|
| 52 |
|
| 53 |
signer_video_fps = signer_video.get_avg_fps()
|
| 54 |
+
# target_fps = 12
|
| 55 |
+
# stride = max(1, int(round(signer_video_fps / target_fps)))
|
| 56 |
+
stride = 1
|
| 57 |
index_list = list(range(0, len(signer_video), stride))
|
| 58 |
signer_video = signer_video.get_batch(index_list)
|
| 59 |
signer_video = signer_video.asnumpy()
|
|
|
|
| 107 |
|
| 108 |
# input_clip = "/share/data/pals/shester/datasets/openasl/clips_bbox/J-0KHhPS_m4.029676-029733.mp4"
|
| 109 |
# input_clip = "/share/data/pals/shester/inference/recordings/sabrin30fps.mp4"
|
| 110 |
+
input_clip = "/share/data/pals/shester/inference/recordings/sample_sabrina.mp4"
|
| 111 |
processor = SHuBERTProcessor(config)
|
| 112 |
output_text = processor.process_video(input_clip)
|
| 113 |
print(f"The English translation is: {output_text}")
|