Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -4,122 +4,108 @@ import gradio as gr
|
|
4 |
import cv2
|
5 |
import time
|
6 |
import os
|
7 |
-
import onnxruntime
|
8 |
from pathlib import Path
|
9 |
-
from ultralytics import YOLO
|
10 |
|
11 |
-
#
|
12 |
-
|
13 |
|
14 |
-
# Set device
|
15 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
16 |
-
|
17 |
-
model.eval()
|
18 |
|
19 |
-
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
-
#
|
23 |
-
|
24 |
-
|
|
|
25 |
|
26 |
-
|
27 |
-
model
|
28 |
-
|
29 |
-
|
30 |
-
export_params=True,
|
31 |
-
opset_version=11,
|
32 |
-
do_constant_folding=True,
|
33 |
-
input_names=["images"],
|
34 |
-
output_names=["output"],
|
35 |
-
dynamic_axes={"images": {0: "batch_size"}, "output": {0: "batch_size"}}
|
36 |
-
)
|
37 |
|
38 |
-
|
39 |
-
session = onnxruntime.InferenceSession(str(model_path), providers=['CUDAExecutionProvider'])
|
40 |
|
41 |
-
# Generate random colors for each class
|
42 |
np.random.seed(42)
|
43 |
-
colors = np.random.uniform(0, 255, size=(
|
44 |
|
45 |
total_inference_time = 0
|
46 |
inference_count = 0
|
47 |
|
48 |
def detect_objects(image):
|
49 |
global total_inference_time, inference_count
|
|
|
50 |
if image is None:
|
51 |
return None
|
52 |
-
|
53 |
-
start_time = time.time()
|
54 |
-
|
55 |
-
# Preprocess image
|
56 |
-
original_shape = image.shape
|
57 |
-
input_shape = (640, 640)
|
58 |
-
image_resized = cv2.resize(image, input_shape)
|
59 |
-
image_norm = image_resized.astype(np.float32) / 255.0
|
60 |
-
image_transposed = np.transpose(image_norm, (2, 0, 1))
|
61 |
-
image_batch = np.expand_dims(image_transposed, axis=0)
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
|
67 |
-
|
68 |
-
|
69 |
|
70 |
-
# Calculate timing
|
71 |
inference_time = time.time() - start_time
|
72 |
total_inference_time += inference_time
|
73 |
inference_count += 1
|
74 |
avg_inference_time = total_inference_time / inference_count
|
75 |
-
fps = 1 / inference_time
|
76 |
|
77 |
-
|
78 |
-
output_image = image.copy()
|
79 |
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
x1, y1, x2, y2,
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
x1, x2 = int(x1 * scale_x), int(x2 * scale_x)
|
92 |
-
y1, y2 = int(y1 * scale_y), int(y2 * scale_y)
|
93 |
-
class_id = int(class_id)
|
94 |
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
label = f"Class {class_id} {conf:.2f}"
|
99 |
-
cv2.putText(output_image, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
|
100 |
|
101 |
-
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
return output_image
|
106 |
|
107 |
-
# Gradio Interface
|
108 |
example_images = ["spring_street_after.jpg", "pexels-hikaique-109919.jpg"]
|
109 |
os.makedirs("examples", exist_ok=True)
|
110 |
|
111 |
with gr.Blocks(title="Optimized YOLOv5 Object Detection") as demo:
|
112 |
-
gr.Markdown("
|
|
|
|
|
|
|
113 |
|
114 |
with gr.Row():
|
115 |
with gr.Column(scale=1):
|
116 |
input_image = gr.Image(label="Input Image", type="numpy")
|
117 |
-
submit_button = gr.Button("
|
118 |
clear_button = gr.Button("Clear")
|
119 |
-
|
120 |
with gr.Column(scale=1):
|
121 |
output_image = gr.Image(label="Detected Objects", type="numpy")
|
122 |
-
|
123 |
gr.Examples(
|
124 |
examples=example_images,
|
125 |
inputs=input_image,
|
@@ -127,7 +113,7 @@ with gr.Blocks(title="Optimized YOLOv5 Object Detection") as demo:
|
|
127 |
fn=detect_objects,
|
128 |
cache_examples=True
|
129 |
)
|
130 |
-
|
131 |
submit_button.click(fn=detect_objects, inputs=input_image, outputs=output_image)
|
132 |
clear_button.click(lambda: (None, None), None, [input_image, output_image])
|
133 |
|
|
|
4 |
import cv2
|
5 |
import time
|
6 |
import os
|
|
|
7 |
from pathlib import Path
|
|
|
8 |
|
9 |
+
# Create cache directory for models
|
10 |
+
os.makedirs("models", exist_ok=True)
|
11 |
|
|
|
12 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
13 |
+
print(f"Using device: {device}")
|
|
|
14 |
|
15 |
+
model_path = Path("models/yolov5n.pt")
|
16 |
+
if model_path.exists():
|
17 |
+
print(f"Loading model from cache: {model_path}")
|
18 |
+
model = torch.hub.load("ultralytics/yolov5", "yolov5x", pretrained=True, source="local", path=str(model_path)).to(device)
|
19 |
+
else:
|
20 |
+
print("Downloading YOLOv5n model and caching...")
|
21 |
+
model = torch.hub.load("ultralytics/yolov5", "yolov5x", pretrained=True).to(device)
|
22 |
+
torch.save(model.state_dict(), model_path)
|
23 |
|
24 |
+
# Model configurations
|
25 |
+
model.conf = 0.3
|
26 |
+
model.iou = 0.3
|
27 |
+
model.classes = None
|
28 |
|
29 |
+
if device.type == "cuda":
|
30 |
+
model.half()
|
31 |
+
else:
|
32 |
+
torch.set_num_threads(os.cpu_count())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
+
model.eval()
|
|
|
35 |
|
|
|
36 |
np.random.seed(42)
|
37 |
+
colors = np.random.uniform(0, 255, size=(len(model.names), 3))
|
38 |
|
39 |
total_inference_time = 0
|
40 |
inference_count = 0
|
41 |
|
42 |
def detect_objects(image):
|
43 |
global total_inference_time, inference_count
|
44 |
+
|
45 |
if image is None:
|
46 |
return None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
+
start_time = time.time()
|
49 |
+
output_image = image.copy()
|
50 |
+
input_size = 640
|
51 |
|
52 |
+
with torch.no_grad():
|
53 |
+
results = model(image, size=input_size)
|
54 |
|
|
|
55 |
inference_time = time.time() - start_time
|
56 |
total_inference_time += inference_time
|
57 |
inference_count += 1
|
58 |
avg_inference_time = total_inference_time / inference_count
|
|
|
59 |
|
60 |
+
detections = results.pred[0].cpu().numpy()
|
|
|
61 |
|
62 |
+
for *xyxy, conf, cls in detections:
|
63 |
+
x1, y1, x2, y2 = map(int, xyxy)
|
64 |
+
class_id = int(cls)
|
65 |
+
color = colors[class_id].tolist()
|
66 |
+
|
67 |
+
# Thicker bounding boxes
|
68 |
+
cv2.rectangle(output_image, (x1, y1), (x2, y2), color, 3, lineType=cv2.LINE_AA)
|
69 |
+
|
70 |
+
label = f"{model.names[class_id]} {conf:.2f}"
|
71 |
+
font_scale, font_thickness = 0.9, 2
|
72 |
+
(w, h), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, font_scale, font_thickness)
|
|
|
|
|
|
|
73 |
|
74 |
+
cv2.rectangle(output_image, (x1, y1 - h - 10), (x1 + w + 10, y1), color, -1)
|
75 |
+
cv2.putText(output_image, label, (x1 + 5, y1 - 5),
|
76 |
+
cv2.FONT_HERSHEY_SIMPLEX, font_scale, (255, 255, 255), font_thickness, lineType=cv2.LINE_AA)
|
|
|
|
|
77 |
|
78 |
+
fps = 1 / inference_time
|
79 |
+
|
80 |
+
# Stylish FPS display
|
81 |
+
overlay = output_image.copy()
|
82 |
+
cv2.rectangle(overlay, (10, 10), (300, 80), (0, 0, 0), -1)
|
83 |
+
output_image = cv2.addWeighted(overlay, 0.6, output_image, 0.4, 0)
|
84 |
+
cv2.putText(output_image, f"FPS: {fps:.2f}", (20, 40),
|
85 |
+
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2, lineType=cv2.LINE_AA)
|
86 |
+
cv2.putText(output_image, f"Avg FPS: {1/avg_inference_time:.2f}", (20, 70),
|
87 |
+
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2, lineType=cv2.LINE_AA)
|
88 |
|
89 |
return output_image
|
90 |
|
|
|
91 |
example_images = ["spring_street_after.jpg", "pexels-hikaique-109919.jpg"]
|
92 |
os.makedirs("examples", exist_ok=True)
|
93 |
|
94 |
with gr.Blocks(title="Optimized YOLOv5 Object Detection") as demo:
|
95 |
+
gr.Markdown("""
|
96 |
+
# Optimized YOLOv5 Object Detection
|
97 |
+
Detects objects using YOLOv5 with enhanced visualization and FPS tracking.
|
98 |
+
""")
|
99 |
|
100 |
with gr.Row():
|
101 |
with gr.Column(scale=1):
|
102 |
input_image = gr.Image(label="Input Image", type="numpy")
|
103 |
+
submit_button = gr.Button("Submit", variant="primary")
|
104 |
clear_button = gr.Button("Clear")
|
105 |
+
|
106 |
with gr.Column(scale=1):
|
107 |
output_image = gr.Image(label="Detected Objects", type="numpy")
|
108 |
+
|
109 |
gr.Examples(
|
110 |
examples=example_images,
|
111 |
inputs=input_image,
|
|
|
113 |
fn=detect_objects,
|
114 |
cache_examples=True
|
115 |
)
|
116 |
+
|
117 |
submit_button.click(fn=detect_objects, inputs=input_image, outputs=output_image)
|
118 |
clear_button.click(lambda: (None, None), None, [input_image, output_image])
|
119 |
|