File size: 6,496 Bytes
1999a98 5f3ef47 acdc36a 5f3ef47 acdc36a d78aa39 7500a70 acdc36a 5f3ef47 7500a70 5f3ef47 7500a70 5f3ef47 4301288 5f3ef47 fde7c40 5f3ef47 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 |
# --------------------------------------------------------
# Based on yolov10
# https://github.com/THU-MIG/yolov10/app.py
# --------------------------------------------------------'
import gradio as gr
import cv2
import tempfile
import spaces, subprocess
# Install flash attention, skipping CUDA build if necessary
subprocess.run(
"pip install flash-attn --no-build-isolation",
env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
shell=True,
)
from ultralytics import YOLO
def results_to_detections(results):
detections = []
for result in results:
boxes = result.boxes.data # [num_det, 6] (x1, y1, x2, y2, conf, cls)
for box in boxes:
x1, y1, x2, y2, conf, cls = box.tolist()
detections.append({
"class_id": int(cls),
"class_name": result.names[int(cls)],
"confidence": float(conf),
"bbox": [x1, y1, x2, y2] # XYXY format
})
return detections
@spaces.GPU
def yolov12_inference(image, video, model_id, image_size, conf_threshold):
model = YOLO(model_id)
if image:
results = model.predict(source=image, imgsz=image_size, conf=conf_threshold)
annotated_image = results[0].plot()
dets = results_to_detections(results)
return annotated_image[:, :, ::-1], None, dets
else:
video_path = tempfile.mktemp(suffix=".webm")
with open(video_path, "wb") as f:
with open(video, "rb") as g:
f.write(g.read())
cap = cv2.VideoCapture(video_path)
fps = cap.get(cv2.CAP_PROP_FPS)
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
output_video_path = tempfile.mktemp(suffix=".webm")
out = cv2.VideoWriter(output_video_path, cv2.VideoWriter_fourcc(*'vp80'), fps, (frame_width, frame_height))
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break
results = model.predict(source=frame, imgsz=image_size, conf=conf_threshold)
annotated_frame = results[0].plot()
out.write(annotated_frame)
cap.release()
out.release()
return None, output_video_path, None
def yolov12_inference_for_examples(image, model_path, image_size, conf_threshold):
annotated_image, _ = yolov12_inference(image, None, model_path, image_size, conf_threshold)
return annotated_image
def app():
with gr.Blocks():
with gr.Row():
with gr.Column():
image = gr.Image(type="pil", label="Image", visible=True)
video = gr.Video(label="Video", visible=False)
input_type = gr.Radio(
choices=["Image", "Video"],
value="Image",
label="Input Type",
)
model_id = gr.Dropdown(
label="Model",
choices=[
"yolov12n.pt",
"yolov12s.pt",
"yolov12m.pt",
"yolov12l.pt",
"yolov12x.pt",
],
value="yolov12m.pt",
)
image_size = gr.Slider(
label="Image Size",
minimum=320,
maximum=1280,
step=32,
value=640,
)
conf_threshold = gr.Slider(
label="Confidence Threshold",
minimum=0.0,
maximum=1.0,
step=0.05,
value=0.25,
)
yolov12_infer = gr.Button(value="Detect Objects")
with gr.Column():
output_image = gr.Image(type="numpy", label="Annotated Image", visible=True)
output_video = gr.Video(label="Annotated Video", visible=False)
def update_visibility(input_type):
image = gr.update(visible=True) if input_type == "Image" else gr.update(visible=False)
video = gr.update(visible=False) if input_type == "Image" else gr.update(visible=True)
output_image = gr.update(visible=True) if input_type == "Image" else gr.update(visible=False)
output_video = gr.update(visible=False) if input_type == "Image" else gr.update(visible=True)
return image, video, output_image, output_video
input_type.change(
fn=update_visibility,
inputs=[input_type],
outputs=[image, video, output_image, output_video],
)
def run_inference(image, video, model_id, image_size, conf_threshold, input_type):
if input_type == "Image":
return yolov12_inference(image, None, model_id, image_size, conf_threshold)
else:
return yolov12_inference(None, video, model_id, image_size, conf_threshold)
yolov12_infer.click(
fn=run_inference,
inputs=[image, video, model_id, image_size, conf_threshold, input_type],
outputs=[output_image, output_video, gr.JSON(visible = False)],
)
gr.Examples(
examples=[
[
"ultralytics/assets/bus.jpg",
"yolov12s.pt",
640,
0.25,
],
[
"ultralytics/assets/zidane.jpg",
"yolov12x.pt",
640,
0.25,
],
],
fn=yolov12_inference_for_examples,
inputs=[
image,
model_id,
image_size,
conf_threshold,
],
outputs=[output_image],
cache_examples='lazy',
)
gradio_app = gr.Blocks()
with gradio_app:
gr.HTML(
"""
<h1 style='text-align: center'>
YOLOv12: Attention-Centric Real-Time Object Detectors
</h1>
""")
gr.HTML(
"""
<h3 style='text-align: center'>
<a href='https://arxiv.org/abs/2502.12524' target='_blank'>arXiv</a> | <a href='https://github.com/sunsmarterjie/yolov12' target='_blank'>github</a>
</h3>
""")
with gr.Row():
with gr.Column():
app()
if __name__ == '__main__':
gradio_app.launch()
|