CodeJackR
commited on
Commit
·
d05bd8d
1
Parent(s):
2f7cfdc
Fix errors
Browse files- handler.py +46 -12
handler.py
CHANGED
@@ -1,41 +1,75 @@
|
|
1 |
# handler.py
|
2 |
|
3 |
import io
|
|
|
4 |
import numpy as np
|
5 |
from PIL import Image
|
6 |
from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
|
7 |
-
from
|
8 |
|
9 |
-
class EndpointHandler(
|
10 |
-
def __init__(self,
|
11 |
"""
|
12 |
Called once at startup.
|
13 |
-
The model files are mounted under /
|
14 |
"""
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
sam = sam_model_registry["vit_b"](checkpoint=checkpoint)
|
18 |
self.mask_generator = SamAutomaticMaskGenerator(sam)
|
19 |
|
20 |
-
|
21 |
"""
|
22 |
Called on every HTTP request.
|
23 |
-
Expecting
|
24 |
"""
|
25 |
-
|
26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
|
|
28 |
img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
|
29 |
img_np = np.array(img)
|
30 |
|
|
|
31 |
masks = self.mask_generator.generate(img_np)
|
32 |
combined = np.zeros(img_np.shape[:2], dtype=np.uint8)
|
33 |
for m in masks:
|
34 |
combined[m["segmentation"]] = 255
|
35 |
|
|
|
36 |
out = io.BytesIO()
|
37 |
Image.fromarray(combined).save(out, format="PNG")
|
38 |
out.seek(0)
|
|
|
39 |
|
40 |
-
# Return
|
41 |
-
return {"
|
|
|
1 |
# handler.py
|
2 |
|
3 |
import io
|
4 |
+
import base64
|
5 |
import numpy as np
|
6 |
from PIL import Image
|
7 |
from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
|
8 |
+
from typing import Dict, List, Any
|
9 |
|
10 |
+
class EndpointHandler():
|
11 |
+
def __init__(self, path=""):
|
12 |
"""
|
13 |
Called once at startup.
|
14 |
+
The model files are mounted under /opt/ml/model by default in Inference Endpoints.
|
15 |
"""
|
16 |
+
# Try different possible checkpoint paths
|
17 |
+
import os
|
18 |
+
possible_paths = [
|
19 |
+
os.path.join(path, "pytorch_model.bin"),
|
20 |
+
os.path.join(path, "model.safetensors"),
|
21 |
+
"/opt/ml/model/pytorch_model.bin",
|
22 |
+
"/opt/ml/model/model.safetensors"
|
23 |
+
]
|
24 |
+
|
25 |
+
checkpoint = None
|
26 |
+
for p in possible_paths:
|
27 |
+
if os.path.exists(p):
|
28 |
+
checkpoint = p
|
29 |
+
break
|
30 |
+
|
31 |
+
if checkpoint is None:
|
32 |
+
raise FileNotFoundError("Could not find model checkpoint in any of the expected locations")
|
33 |
+
|
34 |
sam = sam_model_registry["vit_b"](checkpoint=checkpoint)
|
35 |
self.mask_generator = SamAutomaticMaskGenerator(sam)
|
36 |
|
37 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
38 |
"""
|
39 |
Called on every HTTP request.
|
40 |
+
Expecting base64 encoded image in the 'inputs' field or 'image' field.
|
41 |
"""
|
42 |
+
# Handle different input formats
|
43 |
+
if "inputs" in data:
|
44 |
+
if isinstance(data["inputs"], str):
|
45 |
+
# Base64 encoded image
|
46 |
+
image_bytes = base64.b64decode(data["inputs"])
|
47 |
+
elif isinstance(data["inputs"], dict) and "image" in data["inputs"]:
|
48 |
+
# Nested structure with image field
|
49 |
+
image_bytes = base64.b64decode(data["inputs"]["image"])
|
50 |
+
else:
|
51 |
+
raise ValueError("Invalid input format. Expected base64 encoded image string.")
|
52 |
+
elif "image" in data:
|
53 |
+
# Direct image field
|
54 |
+
image_bytes = base64.b64decode(data["image"])
|
55 |
+
else:
|
56 |
+
raise ValueError("No image found in request. Expected 'inputs' or 'image' field with base64 encoded image.")
|
57 |
|
58 |
+
# Process the image
|
59 |
img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
|
60 |
img_np = np.array(img)
|
61 |
|
62 |
+
# Generate masks
|
63 |
masks = self.mask_generator.generate(img_np)
|
64 |
combined = np.zeros(img_np.shape[:2], dtype=np.uint8)
|
65 |
for m in masks:
|
66 |
combined[m["segmentation"]] = 255
|
67 |
|
68 |
+
# Convert result to base64
|
69 |
out = io.BytesIO()
|
70 |
Image.fromarray(combined).save(out, format="PNG")
|
71 |
out.seek(0)
|
72 |
+
mask_base64 = base64.b64encode(out.getvalue()).decode('utf-8')
|
73 |
|
74 |
+
# Return in the expected format
|
75 |
+
return [{"mask_png_base64": mask_base64, "num_masks": len(masks)}]
|