Spaces:
Running
Running
Add first version of object detection
Browse files
app.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from PIL import Image, ImageDraw
|
3 |
+
from transformers import DetrImageProcessor, DetrForObjectDetection
|
4 |
+
import torch
|
5 |
+
|
6 |
+
# Load DETR model and processor from Hugging Face
|
7 |
+
model_name = "facebook/detr-resnet-50"
|
8 |
+
processor = DetrImageProcessor.from_pretrained(model_name)
|
9 |
+
model = DetrForObjectDetection.from_pretrained(model_name)
|
10 |
+
|
11 |
+
# Main function: takes an image and returns it with boxes and labels
|
12 |
+
def detect_objects(image):
|
13 |
+
inputs = processor(images=image, return_tensors="pt")
|
14 |
+
outputs = model(**inputs)
|
15 |
+
|
16 |
+
# Convert model output to usable detection results
|
17 |
+
target_sizes = torch.tensor([image.size[::-1]])
|
18 |
+
results = processor.post_process_object_detection(
|
19 |
+
outputs, threshold=0.9, target_sizes=target_sizes
|
20 |
+
)[0]
|
21 |
+
|
22 |
+
# Draw bounding boxes and labels on a copy of the image
|
23 |
+
image_with_boxes = image.copy()
|
24 |
+
draw = ImageDraw.Draw(image_with_boxes)
|
25 |
+
|
26 |
+
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
|
27 |
+
box = [round(x, 2) for x in box.tolist()]
|
28 |
+
draw.rectangle(box, outline="red", width=3)
|
29 |
+
label_text = f"{model.config.id2label[label.item()]}: {round(score.item(), 2)}"
|
30 |
+
draw.text((box[0], box[1]), label_text, fill="white")
|
31 |
+
|
32 |
+
return image_with_boxes
|
33 |
+
|
34 |
+
# Gradio interface
|
35 |
+
app = gr.Interface(
|
36 |
+
fn=detect_objects,
|
37 |
+
inputs=gr.Image(type="pil"),
|
38 |
+
outputs=gr.Image()
|
39 |
+
)
|
40 |
+
|
41 |
+
# Run app
|
42 |
+
if __name__ == "__main__":
|
43 |
+
app.launch()
|