raz-1412 commited on
Commit
97c2eaa
·
verified ·
1 Parent(s): 854eb57

Upload 2 files

Browse files
Files changed (2) hide show
  1. app.py +65 -0
  2. requirements.txt +5 -0
app.py ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import torch
3
+ import cv2
4
+ import numpy as np
5
+ import tempfile
6
+ import json
7
+ from transformers import DetrImageProcessor, DetrForObjectDetection
8
+ from PIL import Image
9
+
10
+ # Load a pretrained object detection model
11
+ model_name = "facebook/detr-resnet-50"
12
+ processor = DetrImageProcessor.from_pretrained(model_name)
13
+ model = DetrForObjectDetection.from_pretrained(model_name)
14
+
15
+ # Video processing function
16
+ def process_video(video_path):
17
+ cap = cv2.VideoCapture(video_path)
18
+ frame_rate = cap.get(cv2.CAP_PROP_FPS)
19
+ timestamped_objects = []
20
+
21
+ frame_number = 0
22
+ while cap.isOpened():
23
+ ret, frame = cap.read()
24
+ if not ret:
25
+ break
26
+
27
+ # Process every 10th frame for efficiency
28
+ if frame_number % int(frame_rate) == 0:
29
+ timestamp = frame_number / frame_rate
30
+ pil_image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
31
+
32
+ # Model prediction
33
+ inputs = processor(images=pil_image, return_tensors="pt")
34
+ with torch.no_grad():
35
+ outputs = model(**inputs)
36
+
37
+ # Extract results
38
+ target_sizes = torch.tensor([pil_image.size])
39
+ results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.7)[0]
40
+
41
+ detected_objects = []
42
+ for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
43
+ label_name = model.config.id2label[label.item()]
44
+ detected_objects.append(label_name)
45
+
46
+ if detected_objects:
47
+ timestamped_objects.append({"timestamp": round(timestamp, 2), "objects": list(set(detected_objects))})
48
+
49
+ frame_number += 1
50
+
51
+ cap.release()
52
+ return json.dumps(timestamped_objects, indent=2)
53
+
54
+ # Gradio Interface
55
+ iface = gr.Interface(
56
+ fn=process_video,
57
+ inputs=gr.Video(label="Upload Video"),
58
+ outputs=gr.JSON(label="Timestamped Object Detections"),
59
+ title="Video Object Detector",
60
+ description="Upload a video and get a timestamped list of detected objects using a pretrained DETR model."
61
+ )
62
+
63
+ # Launch the Gradio app
64
+ if __name__ == "__main__":
65
+ iface.launch()
requirements.txt ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ torch
2
+ torchvision
3
+ transformers
4
+ gradio
5
+ opencv-python