Ege Oezsoy
commited on
Commit
•
1c42c71
1
Parent(s):
74033b8
Adjustments
Browse files- endovit_demo.py +21 -8
- endovit_online.py +43 -0
- requirements.txt +2 -1
endovit_demo.py
CHANGED
@@ -5,8 +5,9 @@ from pathlib import Path
|
|
5 |
from timm.models.vision_transformer import VisionTransformer
|
6 |
from functools import partial
|
7 |
from torch import nn
|
|
|
|
|
8 |
|
9 |
-
# requires: pytorch 2.0.1, timm 0.9.16
|
10 |
def process_single_image(image_path, input_size=224, dataset_mean=[0.3464, 0.2280, 0.2228], dataset_std=[0.2520, 0.2128, 0.2093]):
|
11 |
# Define the transformations
|
12 |
transform = T.Compose([
|
@@ -22,18 +23,30 @@ def process_single_image(image_path, input_size=224, dataset_mean=[0.3464, 0.228
|
|
22 |
processed_image = transform(image)
|
23 |
|
24 |
return processed_image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
|
|
26 |
|
27 |
-
|
|
|
28 |
images = torch.stack([process_single_image(image_path) for image_path in image_paths])
|
29 |
|
30 |
device = "cuda"
|
31 |
dtype = torch.float16
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
model = VisionTransformer(patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6)).to(device, dtype).eval()
|
36 |
-
loading = model.load_state_dict(model_weights, strict=False)
|
37 |
-
print(loading)
|
38 |
output = model.forward_features(images.to(device, dtype))
|
39 |
print(output.shape)
|
|
|
5 |
from timm.models.vision_transformer import VisionTransformer
|
6 |
from functools import partial
|
7 |
from torch import nn
|
8 |
+
from huggingface_hub import snapshot_download
|
9 |
+
|
10 |
|
|
|
11 |
def process_single_image(image_path, input_size=224, dataset_mean=[0.3464, 0.2280, 0.2228], dataset_std=[0.2520, 0.2128, 0.2093]):
|
12 |
# Define the transformations
|
13 |
transform = T.Compose([
|
|
|
23 |
processed_image = transform(image)
|
24 |
|
25 |
return processed_image
|
26 |
+
def load_model_from_huggingface(repo_id, model_filename):
|
27 |
+
# Download model files
|
28 |
+
model_path = snapshot_download(repo_id=repo_id, revision="main")
|
29 |
+
model_weights_path = Path(model_path) / model_filename
|
30 |
+
|
31 |
+
# Load model weights
|
32 |
+
model_weights = torch.load(model_weights_path)['model']
|
33 |
+
|
34 |
+
# Define the model (ensure this matches your model's architecture)
|
35 |
+
model = VisionTransformer(patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6)).eval()
|
36 |
+
|
37 |
+
# Load the weights into the model
|
38 |
+
loading = model.load_state_dict(model_weights, strict=False)
|
39 |
|
40 |
+
return model, loading
|
41 |
|
42 |
+
|
43 |
+
image_paths = sorted(Path('demo_images').glob('*.png')) # TODO replace with image pass
|
44 |
images = torch.stack([process_single_image(image_path) for image_path in image_paths])
|
45 |
|
46 |
device = "cuda"
|
47 |
dtype = torch.float16
|
48 |
+
model, loading_info = load_model_from_huggingface("egeozsoy/EndoViT", "endovit.pth")
|
49 |
+
model = model.to(device, dtype)
|
50 |
+
print(loading_info)
|
|
|
|
|
|
|
51 |
output = model.forward_features(images.to(device, dtype))
|
52 |
print(output.shape)
|
endovit_online.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from pathlib import Path
|
3 |
+
from timm.models.vision_transformer import VisionTransformer
|
4 |
+
from functools import partial
|
5 |
+
from torch import nn
|
6 |
+
from huggingface_hub import snapshot_download
|
7 |
+
|
8 |
+
def load_model_from_huggingface(repo_id, model_filename):
|
9 |
+
# Download model files
|
10 |
+
model_path = snapshot_download(repo_id=repo_id, revision="main")
|
11 |
+
model_weights_path = Path(model_path) / model_filename
|
12 |
+
|
13 |
+
# Load model weights
|
14 |
+
model_weights = torch.load(model_weights_path)['model']
|
15 |
+
|
16 |
+
# Define the model (ensure this matches your model's architecture)
|
17 |
+
model = VisionTransformer(patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6)).eval()
|
18 |
+
|
19 |
+
# Load the weights into the model
|
20 |
+
loading = model.load_state_dict(model_weights, strict=False)
|
21 |
+
|
22 |
+
return model, loading
|
23 |
+
def process_single_image(image_path, input_size=224, dataset_mean=[0.3464, 0.2280, 0.2228], dataset_std=[0.2520, 0.2128, 0.2093]):
|
24 |
+
# Define the transformations
|
25 |
+
transform = T.Compose([
|
26 |
+
T.Resize((input_size, input_size)),
|
27 |
+
T.ToTensor(),
|
28 |
+
T.Normalize(mean=dataset_mean, std=dataset_std)
|
29 |
+
])
|
30 |
+
|
31 |
+
# Open the image
|
32 |
+
image = Image.open(image_path).convert('RGB')
|
33 |
+
|
34 |
+
# Apply the transformations
|
35 |
+
processed_image = transform(image)
|
36 |
+
|
37 |
+
return processed_image
|
38 |
+
|
39 |
+
device = "cuda"
|
40 |
+
dtype = torch.float16
|
41 |
+
model, loading_info = load_model_from_huggingface("egeozsoy/EndoViT", "endovit.pth")
|
42 |
+
model = model.to(device, dtype)
|
43 |
+
print(loading_info)
|
requirements.txt
CHANGED
@@ -1,2 +1,3 @@
|
|
1 |
torch==2.0.1
|
2 |
-
timm==0.9.16
|
|
|
|
1 |
torch==2.0.1
|
2 |
+
timm==0.9.16
|
3 |
+
huggingface-hub==0.22.2
|