Update handler.py
Browse files- handler.py +40 -16
handler.py
CHANGED
@@ -1,26 +1,50 @@
|
|
1 |
-
|
2 |
-
|
|
|
|
|
|
|
3 |
import torch
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
-
|
|
|
6 |
def __init__(self, path=""):
|
7 |
-
|
8 |
-
|
|
|
|
|
9 |
|
10 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
11 |
-
self.model.to(device)
|
12 |
-
|
13 |
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
14 |
inputs = data.pop("inputs", data)
|
15 |
-
image_input = inputs.get(
|
16 |
|
|
|
|
|
17 |
|
18 |
-
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
-
# run prediction
|
22 |
generated_ids = self.model.generate(pixel_values.to(device))
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
|
|
|
1 |
+
import base64
|
2 |
+
import io
|
3 |
+
from typing import Any, Dict, List
|
4 |
+
|
5 |
+
import requests
|
6 |
import torch
|
7 |
+
from PIL import Image
|
8 |
+
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
|
9 |
+
|
10 |
+
|
11 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
12 |
|
13 |
+
|
14 |
+
class EndpointHandler:
|
15 |
def __init__(self, path=""):
|
16 |
+
self.processor = TrOCRProcessor.from_pretrained(path)
|
17 |
+
self.model = VisionEncoderDecoderModel.from_pretrained(path)
|
18 |
+
|
19 |
+
self.model.to(device)
|
20 |
|
|
|
|
|
|
|
21 |
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
22 |
inputs = data.pop("inputs", data)
|
23 |
+
image_input = inputs.get("image")
|
24 |
|
25 |
+
if not image_input:
|
26 |
+
return {"error": "No image provided."}
|
27 |
|
28 |
+
try:
|
29 |
+
if image_input.startswith("http"):
|
30 |
+
response = requests.get(image_input, stream=True)
|
31 |
+
if response.status_code == 200:
|
32 |
+
image = Image.open(response.raw).convert("RGB")
|
33 |
+
else:
|
34 |
+
return {
|
35 |
+
"error": f"Failed to fetch image. Status code: {response.status_code}"
|
36 |
+
}
|
37 |
+
else:
|
38 |
+
image_data = base64.b64decode(image_input)
|
39 |
+
image = Image.open(io.BytesIO(image_data)).convert("RGB")
|
40 |
+
except Exception as e:
|
41 |
+
return {"error": f"Failed to process the image. Details: {str(e)}"}
|
42 |
+
|
43 |
+
pixel_values = self.processor(images=image, return_tensors="pt").pixel_values
|
44 |
|
|
|
45 |
generated_ids = self.model.generate(pixel_values.to(device))
|
46 |
+
|
47 |
+
prediction = self.processor.batch_decode(
|
48 |
+
generated_ids, skip_special_tokens=True
|
49 |
+
)
|
50 |
+
return {"text": prediction[0]}
|