Spaces:
Running
on
T4
Running
on
T4
File size: 11,071 Bytes
f590b07 d336c38 30841fb f590b07 3c9b48f d336c38 f590b07 d336c38 39840e5 f590b07 3c9b48f 39840e5 d336c38 3c9b48f f590b07 3c9b48f a23d91c f590b07 d336c38 30841fb d336c38 a23d91c d336c38 30841fb 39840e5 f590b07 39840e5 f590b07 d336c38 f590b07 d336c38 39840e5 3c9b48f 39840e5 f590b07 39840e5 30841fb 3c9b48f d336c38 39840e5 d336c38 3c9b48f d336c38 f590b07 3c9b48f f590b07 3c9b48f d336c38 3c9b48f d336c38 f590b07 d336c38 30841fb 3c9b48f d336c38 3c9b48f 30841fb d336c38 3c9b48f d336c38 3c9b48f d336c38 f590b07 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 |
from typing import List
import os
import cv2
import gradio as gr
import numpy as np
import supervision as sv
import torch
from tqdm import tqdm
from inference.models import YOLOWorld
from utils.efficient_sam import load, inference_with_boxes
from utils.video import generate_file_name, calculate_end_frame_index, create_directory
MARKDOWN = """
# YOLO-World + EfficientSAM 🔥
This is a demo of zero-shot object detection and instance segmentation using
[YOLO-World](https://github.com/AILab-CVC/YOLO-World) and
[EfficientSAM](https://github.com/yformer/EfficientSAM).
Powered by Roboflow [Inference](https://github.com/roboflow/inference) and
[Supervision](https://github.com/roboflow/supervision).
❗ **Don't give up right away if YOLO-World doesn't detect the objects you are looking
for on the first try.** Use the `Configuration` tab and experiment with
`confidence_threshold` and `iou_threshold`. YOLO-World tends to return low `confidence`
values for objects outside the
[COCO](https://universe.roboflow.com/microsoft/coco) dataset. Check out this
[notebook](https://supervision.roboflow.com/develop/notebooks/zero-shot-object-detection-with-yolo-world)
to learn more about YOLO-World's prompting.
"""
RESULTS = "results"
IMAGE_EXAMPLES = [
['https://media.roboflow.com/dog.jpeg', 'dog, eye, nose, tongue, car', 0.005, 0.1, True, False, False],
]
VIDEO_EXAMPLES = [
['https://media.roboflow.com/supervision/video-examples/croissant-1280x720.mp4', 'croissant', 0.01, 0.2, False, False, False],
['https://media.roboflow.com/supervision/video-examples/suitcases-1280x720.mp4', 'suitcase', 0.1, 0.2, False, False, False],
['https://media.roboflow.com/supervision/video-examples/tokyo-walk-1280x720.mp4', 'woman walking', 0.1, 0.2, False, False, False],
['https://media.roboflow.com/supervision/video-examples/wooly-mammoth-1280x720.mp4', 'mammoth', 0.01, 0.2, False, False, False],
]
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
EFFICIENT_SAM_MODEL = load(device=DEVICE)
YOLO_WORLD_MODEL = YOLOWorld(model_id="yolo_world/l")
BOUNDING_BOX_ANNOTATOR = sv.BoundingBoxAnnotator()
MASK_ANNOTATOR = sv.MaskAnnotator()
LABEL_ANNOTATOR = sv.LabelAnnotator()
create_directory(directory_path=RESULTS)
def process_categories(categories: str) -> List[str]:
return [category.strip() for category in categories.split(',')]
def annotate_image(
input_image: np.ndarray,
detections: sv.Detections,
categories: List[str],
with_confidence: bool = False,
) -> np.ndarray:
labels = [
(
f"{categories[class_id]}: {confidence:.3f}"
if with_confidence
else f"{categories[class_id]}"
)
for class_id, confidence in
zip(detections.class_id, detections.confidence)
]
output_image = MASK_ANNOTATOR.annotate(input_image, detections)
output_image = BOUNDING_BOX_ANNOTATOR.annotate(output_image, detections)
output_image = LABEL_ANNOTATOR.annotate(output_image, detections, labels=labels)
return output_image
def process_image(
input_image: np.ndarray,
categories: str,
confidence_threshold: float = 0.3,
iou_threshold: float = 0.5,
with_segmentation: bool = True,
with_confidence: bool = False,
with_class_agnostic_nms: bool = False,
) -> np.ndarray:
categories = process_categories(categories)
YOLO_WORLD_MODEL.set_classes(categories)
results = YOLO_WORLD_MODEL.infer(input_image, confidence=confidence_threshold)
detections = sv.Detections.from_inference(results)
detections = detections.with_nms(
class_agnostic=with_class_agnostic_nms,
threshold=iou_threshold
)
if with_segmentation:
detections.mask = inference_with_boxes(
image=input_image,
xyxy=detections.xyxy,
model=EFFICIENT_SAM_MODEL,
device=DEVICE
)
output_image = cv2.cvtColor(input_image, cv2.COLOR_RGB2BGR)
output_image = annotate_image(
input_image=output_image,
detections=detections,
categories=categories,
with_confidence=with_confidence
)
return cv2.cvtColor(output_image, cv2.COLOR_BGR2RGB)
def process_video(
input_video: str,
categories: str,
confidence_threshold: float = 0.3,
iou_threshold: float = 0.5,
with_segmentation: bool = True,
with_confidence: bool = False,
with_class_agnostic_nms: bool = False,
progress=gr.Progress(track_tqdm=True)
) -> str:
categories = process_categories(categories)
YOLO_WORLD_MODEL.set_classes(categories)
video_info = sv.VideoInfo.from_video_path(input_video)
total = calculate_end_frame_index(input_video)
frame_generator = sv.get_video_frames_generator(
source_path=input_video,
end=total
)
result_file_name = generate_file_name(extension="mp4")
result_file_path = os.path.join(RESULTS, result_file_name)
with sv.VideoSink(result_file_path, video_info=video_info) as sink:
for _ in tqdm(range(total), desc="Processing video..."):
frame = next(frame_generator)
results = YOLO_WORLD_MODEL.infer(frame, confidence=confidence_threshold)
detections = sv.Detections.from_inference(results)
detections = detections.with_nms(
class_agnostic=with_class_agnostic_nms,
threshold=iou_threshold
)
if with_segmentation:
detections.mask = inference_with_boxes(
image=frame,
xyxy=detections.xyxy,
model=EFFICIENT_SAM_MODEL,
device=DEVICE
)
frame = annotate_image(
input_image=frame,
detections=detections,
categories=categories,
with_confidence=with_confidence
)
sink.write_frame(frame)
return result_file_path
confidence_threshold_component = gr.Slider(
minimum=0,
maximum=1.0,
value=0.3,
step=0.01,
label="Confidence Threshold",
info=(
"The confidence threshold for the YOLO-World model. Lower the threshold to "
"reduce false negatives, enhancing the model's sensitivity to detect "
"sought-after objects. Conversely, increase the threshold to minimize false "
"positives, preventing the model from identifying objects it shouldn't."
))
iou_threshold_component = gr.Slider(
minimum=0,
maximum=1.0,
value=0.5,
step=0.01,
label="IoU Threshold",
info=(
"The Intersection over Union (IoU) threshold for non-maximum suppression. "
"Decrease the value to lessen the occurrence of overlapping bounding boxes, "
"making the detection process stricter. On the other hand, increase the value "
"to allow more overlapping bounding boxes, accommodating a broader range of "
"detections."
))
with_segmentation_component = gr.Checkbox(
value=True,
label="With Segmentation",
info=(
"Whether to run EfficientSAM for instance segmentation."
)
)
with_confidence_component = gr.Checkbox(
value=False,
label="Display Confidence",
info=(
"Whether to display the confidence of the detected objects."
)
)
with_class_agnostic_nms_component = gr.Checkbox(
value=False,
label="Use Class-Agnostic NMS",
info=(
"Suppress overlapping bounding boxes across all classes."
)
)
with gr.Blocks() as demo:
gr.Markdown(MARKDOWN)
with gr.Accordion("Configuration", open=False):
confidence_threshold_component.render()
iou_threshold_component.render()
with gr.Row():
with_segmentation_component.render()
with_confidence_component.render()
with_class_agnostic_nms_component.render()
with gr.Tab(label="Image"):
with gr.Row():
input_image_component = gr.Image(
type='numpy',
label='Input Image'
)
output_image_component = gr.Image(
type='numpy',
label='Output Image'
)
with gr.Row():
image_categories_text_component = gr.Textbox(
label='Categories',
placeholder='comma separated list of categories',
scale=7
)
image_submit_button_component = gr.Button(
value='Submit',
scale=1,
variant='primary'
)
gr.Examples(
fn=process_image,
examples=IMAGE_EXAMPLES,
inputs=[
input_image_component,
image_categories_text_component,
confidence_threshold_component,
iou_threshold_component,
with_segmentation_component,
with_confidence_component,
with_class_agnostic_nms_component
],
outputs=output_image_component
)
with gr.Tab(label="Video"):
with gr.Row():
input_video_component = gr.Video(
label='Input Video'
)
output_video_component = gr.Video(
label='Output Video'
)
with gr.Row():
video_categories_text_component = gr.Textbox(
label='Categories',
placeholder='comma separated list of categories',
scale=7
)
video_submit_button_component = gr.Button(
value='Submit',
scale=1,
variant='primary'
)
gr.Examples(
fn=process_video,
examples=VIDEO_EXAMPLES,
inputs=[
input_video_component,
video_categories_text_component,
confidence_threshold_component,
iou_threshold_component,
with_segmentation_component,
with_confidence_component,
with_class_agnostic_nms_component
],
outputs=output_image_component
)
image_submit_button_component.click(
fn=process_image,
inputs=[
input_image_component,
image_categories_text_component,
confidence_threshold_component,
iou_threshold_component,
with_segmentation_component,
with_confidence_component,
with_class_agnostic_nms_component
],
outputs=output_image_component
)
video_submit_button_component.click(
fn=process_video,
inputs=[
input_video_component,
video_categories_text_component,
confidence_threshold_component,
iou_threshold_component,
with_segmentation_component,
with_confidence_component,
with_class_agnostic_nms_component
],
outputs=output_video_component
)
demo.launch(debug=False, show_error=True)
|