Komal133 commited on
Commit
581bbb4
·
verified ·
1 Parent(s): 41d4208

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +35 -31
app.py CHANGED
@@ -1,22 +1,26 @@
1
  import gradio as gr
2
  from PIL import Image, ImageDraw
3
  import torch
4
- from torchvision import transforms
5
- import torchvision
6
- import io
7
 
8
- # Load your custom-trained model (replace with actual path)
9
- model = torch.load("model.pth", map_location=torch.device('cpu'))
10
  model.eval()
11
 
12
- # Your actual defect classes
13
- class_labels = ["crack", "spalling", "rust", "deformation"]
14
- severity_mapping = {
15
- "crack": "medium",
16
- "spalling": "high",
17
- "rust": "medium",
18
- "deformation": "high"
19
- }
 
 
 
 
 
 
20
 
21
  transform = transforms.Compose([
22
  transforms.ToTensor(),
@@ -26,29 +30,29 @@ def detect_defects(image):
26
  image_tensor = transform(image).unsqueeze(0)
27
 
28
  with torch.no_grad():
29
- predictions = model(image_tensor)[0]
30
 
31
  result_image = image.copy()
32
  draw = ImageDraw.Draw(result_image)
33
- output = []
34
 
35
- for box, label, score in zip(predictions['boxes'], predictions['labels'], predictions['scores']):
36
- score = score.item()
37
- label = label.item()
 
 
38
 
39
- if score > 0.3 and label < len(class_labels): # Filter by confidence and known classes
40
- box = [int(x) for x in box.tolist()]
41
- defect_type = class_labels[label]
42
- severity = severity_mapping.get(defect_type, "low")
43
 
44
- draw.rectangle(box, outline="red", width=3)
45
- draw.text((box[0], box[1] - 10), f"{defect_type} ({score:.2f})", fill="red")
 
 
 
46
 
47
- output.append({
48
- "type": defect_type,
49
- "confidence": round(score, 2),
50
- "severity": severity
51
- })
52
 
53
  return result_image, output
54
 
@@ -59,8 +63,8 @@ demo = gr.Interface(
59
  gr.Image(label="Detection Result"),
60
  gr.JSON(label="Detected Faults with Severity")
61
  ],
62
- title="Structural Defect Detection",
63
- description="Upload a high-res image captured by a drone to detect structural faults such as cracks, rust, and spalling."
64
  )
65
 
66
  if __name__ == "__main__":
 
1
  import gradio as gr
2
  from PIL import Image, ImageDraw
3
  import torch
4
+ from torchvision import models, transforms
 
 
5
 
6
+ # ✅ Directly load a pretrained model (no model.pth needed)
7
+ model = models.detection.fasterrcnn_resnet50_fpn(pretrained=True)
8
  model.eval()
9
 
10
+ # You must define default labels (COCO classes)
11
+ COCO_INSTANCE_CATEGORY_NAMES = [
12
+ '__background__', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
13
+ 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter',
14
+ 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra',
15
+ 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis',
16
+ 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard',
17
+ 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon',
18
+ 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza',
19
+ 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet',
20
+ 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven',
21
+ 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear',
22
+ 'hair drier', 'toothbrush'
23
+ ]
24
 
25
  transform = transforms.Compose([
26
  transforms.ToTensor(),
 
30
  image_tensor = transform(image).unsqueeze(0)
31
 
32
  with torch.no_grad():
33
+ predictions = model(image_tensor)
34
 
35
  result_image = image.copy()
36
  draw = ImageDraw.Draw(result_image)
 
37
 
38
+ output = []
39
+ for i in range(len(predictions[0]['boxes'])):
40
+ score = predictions[0]['scores'][i].item()
41
+ if score < 0.7: # filter low-confidence predictions
42
+ continue
43
 
44
+ box = predictions[0]['boxes'][i].tolist()
45
+ label_idx = predictions[0]['labels'][i].item()
46
+ label = COCO_INSTANCE_CATEGORY_NAMES[label_idx]
 
47
 
48
+ output.append({
49
+ "type": label,
50
+ "confidence": round(score, 2),
51
+ "severity": "unknown"
52
+ })
53
 
54
+ draw.rectangle(box, outline="red", width=3)
55
+ draw.text((box[0], box[1]), label, fill="red")
 
 
 
56
 
57
  return result_image, output
58
 
 
63
  gr.Image(label="Detection Result"),
64
  gr.JSON(label="Detected Faults with Severity")
65
  ],
66
+ title="Structural Defect Detection (Demo)",
67
+ description="Detects general objects using Faster R-CNN pretrained on COCO. Not specific to cracks or faults."
68
  )
69
 
70
  if __name__ == "__main__":