Upload folder using huggingface_hub
Browse files- __pycache__/miner.cpython-312.pyc +0 -0
- __pycache__/pitch.cpython-312.pyc +0 -0
- pitch.py +7 -2
__pycache__/miner.cpython-312.pyc
CHANGED
|
Binary files a/__pycache__/miner.cpython-312.pyc and b/__pycache__/miner.cpython-312.pyc differ
|
|
|
__pycache__/pitch.cpython-312.pyc
CHANGED
|
Binary files a/__pycache__/pitch.cpython-312.pyc and b/__pycache__/pitch.cpython-312.pyc differ
|
|
|
pitch.py
CHANGED
|
@@ -16,6 +16,8 @@ import torchvision.transforms as T
|
|
| 16 |
import torchvision.transforms.functional as f
|
| 17 |
from pydantic import BaseModel
|
| 18 |
|
|
|
|
|
|
|
| 19 |
import logging
|
| 20 |
logger = logging.getLogger(__name__)
|
| 21 |
|
|
@@ -614,6 +616,9 @@ def inference_batch(frames, model, kp_threshold, device, batch_size=8):
|
|
| 614 |
model_device = next(model.parameters()).device
|
| 615 |
print(model_device)
|
| 616 |
|
|
|
|
|
|
|
|
|
|
| 617 |
# Process all frames in optimally-sized batches
|
| 618 |
for i in range(0, num_frames, batch_size):
|
| 619 |
current_batch_size = min(batch_size, num_frames - i)
|
|
@@ -626,8 +631,8 @@ def inference_batch(frames, model, kp_threshold, device, batch_size=8):
|
|
| 626 |
# Move batch to model device
|
| 627 |
batch = batch.to(model_device)
|
| 628 |
|
| 629 |
-
with torch.no_grad():
|
| 630 |
-
|
| 631 |
|
| 632 |
# Ultra-fast keypoint extraction
|
| 633 |
kp_coords = extract_keypoints_from_heatmap_fast(heatmaps[:,:-1,:,:], scale=2, max_keypoints=1)
|
|
|
|
| 16 |
import torchvision.transforms.functional as f
|
| 17 |
from pydantic import BaseModel
|
| 18 |
|
| 19 |
+
import tensorflow as tf
|
| 20 |
+
|
| 21 |
import logging
|
| 22 |
logger = logging.getLogger(__name__)
|
| 23 |
|
|
|
|
| 616 |
model_device = next(model.parameters()).device
|
| 617 |
print(model_device)
|
| 618 |
|
| 619 |
+
@tf.function
|
| 620 |
+
def infer(x):
|
| 621 |
+
return model(x)
|
| 622 |
# Process all frames in optimally-sized batches
|
| 623 |
for i in range(0, num_frames, batch_size):
|
| 624 |
current_batch_size = min(batch_size, num_frames - i)
|
|
|
|
| 631 |
# Move batch to model device
|
| 632 |
batch = batch.to(model_device)
|
| 633 |
|
| 634 |
+
# with torch.no_grad():
|
| 635 |
+
heatmaps = infer(batch)
|
| 636 |
|
| 637 |
# Ultra-fast keypoint extraction
|
| 638 |
kp_coords = extract_keypoints_from_heatmap_fast(heatmaps[:,:-1,:,:], scale=2, max_keypoints=1)
|