Create inference.py
Browse files- inference.py +47 -0
inference.py
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torchvision.transforms as T
|
3 |
+
from PIL import Image
|
4 |
+
|
5 |
+
# Load the YOLOv5 model
|
6 |
+
model = torch.hub.load('niki-stha/asl-detection-yolov5', 'yolov5s')
|
7 |
+
|
8 |
+
# Set the device (GPU if available, otherwise CPU)
|
9 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
10 |
+
model.to(device).eval()
|
11 |
+
|
12 |
+
# Define the image transformation
|
13 |
+
transform = T.Compose([
|
14 |
+
T.Resize((416, 416)),
|
15 |
+
T.ToTensor(),
|
16 |
+
])
|
17 |
+
|
18 |
+
# Inference function
|
19 |
+
def run_inference(image):
|
20 |
+
# Preprocess the image
|
21 |
+
image = transform(image).unsqueeze(0).to(device)
|
22 |
+
|
23 |
+
# Perform inference
|
24 |
+
results = model(image)
|
25 |
+
|
26 |
+
# Post-process the results
|
27 |
+
# (You can customize this part based on your specific requirements)
|
28 |
+
predictions = results.pandas().xyxy[0]
|
29 |
+
|
30 |
+
return predictions
|
31 |
+
|
32 |
+
# Example API endpoint
|
33 |
+
def inference_api(request):
|
34 |
+
# Get the image from the request (you may need to adapt this based on your API framework)
|
35 |
+
image_data = request.files['image'].read()
|
36 |
+
image = Image.open(io.BytesIO(image_data))
|
37 |
+
|
38 |
+
# Run inference
|
39 |
+
predictions = run_inference(image)
|
40 |
+
|
41 |
+
# Convert predictions to JSON or any other desired format
|
42 |
+
# (You may need to adapt this based on your API framework)
|
43 |
+
response = {
|
44 |
+
'predictions': predictions.to_dict(orient='records')
|
45 |
+
}
|
46 |
+
|
47 |
+
return jsonify(response)
|