Upload detect.py
Browse files
detect.py
ADDED
@@ -0,0 +1,312 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
|
2 |
+
"""
|
3 |
+
Run YOLOv5 detection inference on images, videos, directories, globs, YouTube, webcam, streams, etc.
|
4 |
+
|
5 |
+
Usage - sources:
|
6 |
+
$ python detect.py --weights yolov5s.pt --source 0 # webcam
|
7 |
+
img.jpg # image
|
8 |
+
vid.mp4 # video
|
9 |
+
screen # screenshot
|
10 |
+
path/ # directory
|
11 |
+
list.txt # list of images
|
12 |
+
list.streams # list of streams
|
13 |
+
'path/*.jpg' # glob
|
14 |
+
'https://youtu.be/LNwODJXcvt4' # YouTube
|
15 |
+
'rtsp://example.com/media.mp4' # RTSP, RTMP, HTTP stream
|
16 |
+
|
17 |
+
Usage - formats:
|
18 |
+
$ python detect.py --weights yolov5s.pt # PyTorch
|
19 |
+
yolov5s.torchscript # TorchScript
|
20 |
+
yolov5s.onnx # ONNX Runtime or OpenCV DNN with --dnn
|
21 |
+
yolov5s_openvino_model # OpenVINO
|
22 |
+
yolov5s.engine # TensorRT
|
23 |
+
yolov5s.mlmodel # CoreML (macOS-only)
|
24 |
+
yolov5s_saved_model # TensorFlow SavedModel
|
25 |
+
yolov5s.pb # TensorFlow GraphDef
|
26 |
+
yolov5s.tflite # TensorFlow Lite
|
27 |
+
yolov5s_edgetpu.tflite # TensorFlow Edge TPU
|
28 |
+
yolov5s_paddle_model # PaddlePaddle
|
29 |
+
"""
|
30 |
+
|
31 |
+
import argparse
|
32 |
+
import csv
|
33 |
+
import os
|
34 |
+
import platform
|
35 |
+
import sys
|
36 |
+
from pathlib import Path
|
37 |
+
|
38 |
+
import torch
|
39 |
+
|
40 |
+
FILE = Path(__file__).resolve()
|
41 |
+
ROOT = FILE.parents[0] # YOLOv5 root directory
|
42 |
+
if str(ROOT) not in sys.path:
|
43 |
+
sys.path.append(str(ROOT)) # add ROOT to PATH
|
44 |
+
ROOT = Path(os.path.relpath(ROOT, Path.cwd())) # relative
|
45 |
+
|
46 |
+
from ultralytics.utils.plotting import Annotator, colors, save_one_box
|
47 |
+
|
48 |
+
from models.common import DetectMultiBackend
|
49 |
+
from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams
|
50 |
+
from utils.general import (
|
51 |
+
LOGGER,
|
52 |
+
Profile,
|
53 |
+
check_file,
|
54 |
+
check_img_size,
|
55 |
+
check_imshow,
|
56 |
+
check_requirements,
|
57 |
+
colorstr,
|
58 |
+
cv2,
|
59 |
+
increment_path,
|
60 |
+
non_max_suppression,
|
61 |
+
print_args,
|
62 |
+
scale_boxes,
|
63 |
+
strip_optimizer,
|
64 |
+
xyxy2xywh,
|
65 |
+
)
|
66 |
+
from utils.torch_utils import select_device, smart_inference_mode
|
67 |
+
|
68 |
+
|
69 |
+
@smart_inference_mode()
|
70 |
+
def run(
|
71 |
+
weights=ROOT / "yolov5s.pt", # model path or triton URL
|
72 |
+
source=ROOT / "data/images", # file/dir/URL/glob/screen/0(webcam)
|
73 |
+
data=ROOT / "data/coco128.yaml", # dataset.yaml path
|
74 |
+
imgsz=(640, 640), # inference size (height, width)
|
75 |
+
conf_thres=0.25, # confidence threshold
|
76 |
+
iou_thres=0.025, # NMS IOU threshold
|
77 |
+
max_det=1000, # maximum detections per image
|
78 |
+
device="", # cuda device, i.e. 0 or 0,1,2,3 or cpu
|
79 |
+
view_img=False, # show results
|
80 |
+
save_txt=False, # save results to *.txt
|
81 |
+
save_csv=False, # save results in CSV format
|
82 |
+
save_conf=False, # save confidences in --save-txt labels
|
83 |
+
save_crop=False, # save cropped prediction boxes
|
84 |
+
nosave=False, # do not save images/videos
|
85 |
+
classes=None, # filter by class: --class 0, or --class 0 2 3
|
86 |
+
agnostic_nms=False, # class-agnostic NMS
|
87 |
+
augment=False, # augmented inference
|
88 |
+
visualize=False, # visualize features
|
89 |
+
update=False, # update all models
|
90 |
+
project=ROOT / "runs/detect", # save results to project/name
|
91 |
+
name="exp", # save results to project/name
|
92 |
+
exist_ok=False, # existing project/name ok, do not increment
|
93 |
+
line_thickness=3, # bounding box thickness (pixels)
|
94 |
+
hide_labels=False, # hide labels
|
95 |
+
hide_conf=False, # hide confidences
|
96 |
+
half=False, # use FP16 half-precision inference
|
97 |
+
dnn=False, # use OpenCV DNN for ONNX inference
|
98 |
+
vid_stride=1, # video frame-rate stride
|
99 |
+
):
|
100 |
+
source = str(source)
|
101 |
+
save_img = not nosave and not source.endswith(".txt") # save inference images
|
102 |
+
is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
|
103 |
+
is_url = source.lower().startswith(("rtsp://", "rtmp://", "http://", "https://"))
|
104 |
+
webcam = source.isnumeric() or source.endswith(".streams") or (is_url and not is_file)
|
105 |
+
screenshot = source.lower().startswith("screen")
|
106 |
+
if is_url and is_file:
|
107 |
+
source = check_file(source) # download
|
108 |
+
|
109 |
+
# Directories
|
110 |
+
save_dir = increment_path(Path(project) / name, exist_ok=exist_ok) # increment run
|
111 |
+
(save_dir / "labels" if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
|
112 |
+
|
113 |
+
# Load model
|
114 |
+
device = select_device(device)
|
115 |
+
model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
|
116 |
+
stride, names, pt = model.stride, model.names, model.pt
|
117 |
+
imgsz = check_img_size(imgsz, s=stride) # check image size
|
118 |
+
|
119 |
+
# Dataloader
|
120 |
+
bs = 1 # batch_size
|
121 |
+
if webcam:
|
122 |
+
view_img = check_imshow(warn=True)
|
123 |
+
dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
|
124 |
+
bs = len(dataset)
|
125 |
+
elif screenshot:
|
126 |
+
dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)
|
127 |
+
else:
|
128 |
+
dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
|
129 |
+
vid_path, vid_writer = [None] * bs, [None] * bs
|
130 |
+
|
131 |
+
# Run inference
|
132 |
+
model.warmup(imgsz=(1 if pt or model.triton else bs, 3, *imgsz)) # warmup
|
133 |
+
seen, windows, dt = 0, [], (Profile(device=device), Profile(device=device), Profile(device=device))
|
134 |
+
for path, im, im0s, vid_cap, s in dataset:
|
135 |
+
with dt[0]:
|
136 |
+
im = torch.from_numpy(im).to(model.device)
|
137 |
+
im = im.half() if model.fp16 else im.float() # uint8 to fp16/32
|
138 |
+
im /= 255 # 0 - 255 to 0.0 - 1.0
|
139 |
+
if len(im.shape) == 3:
|
140 |
+
im = im[None] # expand for batch dim
|
141 |
+
if model.xml and im.shape[0] > 1:
|
142 |
+
ims = torch.chunk(im, im.shape[0], 0)
|
143 |
+
|
144 |
+
# Inference
|
145 |
+
with dt[1]:
|
146 |
+
visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
|
147 |
+
if model.xml and im.shape[0] > 1:
|
148 |
+
pred = None
|
149 |
+
for image in ims:
|
150 |
+
if pred is None:
|
151 |
+
pred = model(image, augment=augment, visualize=visualize).unsqueeze(0)
|
152 |
+
else:
|
153 |
+
pred = torch.cat((pred, model(image, augment=augment, visualize=visualize).unsqueeze(0)), dim=0)
|
154 |
+
pred = [pred, None]
|
155 |
+
else:
|
156 |
+
pred = model(im, augment=augment, visualize=visualize)
|
157 |
+
# NMS
|
158 |
+
with dt[2]:
|
159 |
+
pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)
|
160 |
+
|
161 |
+
# Second-stage classifier (optional)
|
162 |
+
# pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
|
163 |
+
|
164 |
+
# Define the path for the CSV file
|
165 |
+
csv_path = save_dir / "predictions.csv"
|
166 |
+
|
167 |
+
# Create or append to the CSV file
|
168 |
+
def write_to_csv(image_name, prediction, confidence):
|
169 |
+
"""Writes prediction data for an image to a CSV file, appending if the file exists."""
|
170 |
+
data = {"Image Name": image_name, "Prediction": prediction, "Confidence": confidence}
|
171 |
+
with open(csv_path, mode="a", newline="") as f:
|
172 |
+
writer = csv.DictWriter(f, fieldnames=data.keys())
|
173 |
+
if not csv_path.is_file():
|
174 |
+
writer.writeheader()
|
175 |
+
writer.writerow(data)
|
176 |
+
|
177 |
+
# Process predictions
|
178 |
+
for i, det in enumerate(pred): # per image
|
179 |
+
seen += 1
|
180 |
+
if webcam: # batch_size >= 1
|
181 |
+
p, im0, frame = path[i], im0s[i].copy(), dataset.count
|
182 |
+
s += f"{i}: "
|
183 |
+
else:
|
184 |
+
p, im0, frame = path, im0s.copy(), getattr(dataset, "frame", 0)
|
185 |
+
|
186 |
+
p = Path(p) # to Path
|
187 |
+
save_path = str(save_dir / p.name) # im.jpg
|
188 |
+
txt_path = str(save_dir / "labels" / p.stem) + ("" if dataset.mode == "image" else f"_{frame}") # im.txt
|
189 |
+
s += "%gx%g " % im.shape[2:] # print string
|
190 |
+
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh
|
191 |
+
imc = im0.copy() if save_crop else im0 # for save_crop
|
192 |
+
annotator = Annotator(im0, line_width=line_thickness, example=str(names))
|
193 |
+
if len(det):
|
194 |
+
# Rescale boxes from img_size to im0 size
|
195 |
+
det[:, :4] = scale_boxes(im.shape[2:], det[:, :4], im0.shape).round()
|
196 |
+
|
197 |
+
# Print results
|
198 |
+
for c in det[:, 5].unique():
|
199 |
+
n = (det[:, 5] == c).sum() # detections per class
|
200 |
+
s += f"{n} {names[int(c)]}{'s' * (n > 1)}, " # add to string
|
201 |
+
|
202 |
+
# Write results
|
203 |
+
for *xyxy, conf, cls in reversed(det):
|
204 |
+
c = int(cls) # integer class
|
205 |
+
label = names[c] if hide_conf else f"{names[c]}"
|
206 |
+
confidence = float(conf)
|
207 |
+
confidence_str = f"{confidence:.2f}"
|
208 |
+
|
209 |
+
if save_csv:
|
210 |
+
write_to_csv(p.name, label, confidence_str)
|
211 |
+
|
212 |
+
if save_txt: # Write to file
|
213 |
+
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
|
214 |
+
line = (cls, *xywh, conf) if save_conf else (cls, *xywh) # label format
|
215 |
+
with open(f"{txt_path}.txt", "a") as f:
|
216 |
+
f.write(("%g " * len(line)).rstrip() % line + "\n")
|
217 |
+
|
218 |
+
if save_img or save_crop or view_img: # Add bbox to image
|
219 |
+
c = int(cls) # integer class
|
220 |
+
label = None if hide_labels else (names[c] if hide_conf else f"{names[c]} {conf:.2f}")
|
221 |
+
annotator.box_label(xyxy, label, color=colors(c, True))
|
222 |
+
if save_crop:
|
223 |
+
save_one_box(xyxy, imc, file=save_dir / "crops" / names[c] / f"{p.stem}.jpg", BGR=True)
|
224 |
+
|
225 |
+
# Stream results
|
226 |
+
im0 = annotator.result()
|
227 |
+
if view_img:
|
228 |
+
if platform.system() == "Linux" and p not in windows:
|
229 |
+
windows.append(p)
|
230 |
+
cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO) # allow window resize (Linux)
|
231 |
+
cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])
|
232 |
+
cv2.imshow(str(p), im0)
|
233 |
+
cv2.waitKey(1) # 1 millisecond
|
234 |
+
|
235 |
+
# Save results (image with detections)
|
236 |
+
if save_img:
|
237 |
+
if dataset.mode == "image":
|
238 |
+
cv2.imwrite(save_path, im0)
|
239 |
+
else: # 'video' or 'stream'
|
240 |
+
if vid_path[i] != save_path: # new video
|
241 |
+
vid_path[i] = save_path
|
242 |
+
if isinstance(vid_writer[i], cv2.VideoWriter):
|
243 |
+
vid_writer[i].release() # release previous video writer
|
244 |
+
if vid_cap: # video
|
245 |
+
fps = vid_cap.get(cv2.CAP_PROP_FPS)
|
246 |
+
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
247 |
+
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
248 |
+
else: # stream
|
249 |
+
fps, w, h = 30, im0.shape[1], im0.shape[0]
|
250 |
+
save_path = str(Path(save_path).with_suffix(".mp4")) # force *.mp4 suffix on results videos
|
251 |
+
vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*"mp4v"), fps, (w, h))
|
252 |
+
vid_writer[i].write(im0)
|
253 |
+
|
254 |
+
# Print time (inference-only)
|
255 |
+
LOGGER.info(f"{s}{'' if len(det) else '(no detections), '}{dt[1].dt * 1E3:.1f}ms")
|
256 |
+
|
257 |
+
# Print results
|
258 |
+
t = tuple(x.t / seen * 1e3 for x in dt) # speeds per image
|
259 |
+
LOGGER.info(f"Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}" % t)
|
260 |
+
if save_txt or save_img:
|
261 |
+
s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ""
|
262 |
+
LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
|
263 |
+
if update:
|
264 |
+
strip_optimizer(weights[0]) # update model (to fix SourceChangeWarning)
|
265 |
+
|
266 |
+
|
267 |
+
def parse_opt():
|
268 |
+
"""Parses command-line arguments for YOLOv5 detection, setting inference options and model configurations."""
|
269 |
+
parser = argparse.ArgumentParser()
|
270 |
+
parser.add_argument("--weights", nargs="+", type=str, default=ROOT / "yolov5s.pt", help="model path or triton URL")
|
271 |
+
parser.add_argument("--source", type=str, default=ROOT / "data/images", help="file/dir/URL/glob/screen/0(webcam)")
|
272 |
+
parser.add_argument("--data", type=str, default=ROOT / "data/coco128.yaml", help="(optional) dataset.yaml path")
|
273 |
+
parser.add_argument("--imgsz", "--img", "--img-size", nargs="+", type=int, default=[640], help="inference size h,w")
|
274 |
+
parser.add_argument("--conf-thres", type=float, default=0.25, help="confidence threshold")
|
275 |
+
parser.add_argument("--iou-thres", type=float, default=0.45, help="NMS IoU threshold")
|
276 |
+
parser.add_argument("--max-det", type=int, default=1000, help="maximum detections per image")
|
277 |
+
parser.add_argument("--device", default="", help="cuda device, i.e. 0 or 0,1,2,3 or cpu")
|
278 |
+
parser.add_argument("--view-img", action="store_true", help="show results")
|
279 |
+
parser.add_argument("--save-txt", action="store_true", help="save results to *.txt")
|
280 |
+
parser.add_argument("--save-csv", action="store_true", help="save results in CSV format")
|
281 |
+
parser.add_argument("--save-conf", action="store_true", help="save confidences in --save-txt labels")
|
282 |
+
parser.add_argument("--save-crop", action="store_true", help="save cropped prediction boxes")
|
283 |
+
parser.add_argument("--nosave", action="store_true", help="do not save images/videos")
|
284 |
+
parser.add_argument("--classes", nargs="+", type=int, help="filter by class: --classes 0, or --classes 0 2 3")
|
285 |
+
parser.add_argument("--agnostic-nms", action="store_true", help="class-agnostic NMS")
|
286 |
+
parser.add_argument("--augment", action="store_true", help="augmented inference")
|
287 |
+
parser.add_argument("--visualize", action="store_true", help="visualize features")
|
288 |
+
parser.add_argument("--update", action="store_true", help="update all models")
|
289 |
+
parser.add_argument("--project", default=ROOT / "runs/detect", help="save results to project/name")
|
290 |
+
parser.add_argument("--name", default="exp", help="save results to project/name")
|
291 |
+
parser.add_argument("--exist-ok", action="store_true", help="existing project/name ok, do not increment")
|
292 |
+
parser.add_argument("--line-thickness", default=3, type=int, help="bounding box thickness (pixels)")
|
293 |
+
parser.add_argument("--hide-labels", default=False, action="store_true", help="hide labels")
|
294 |
+
parser.add_argument("--hide-conf", default=False, action="store_true", help="hide confidences")
|
295 |
+
parser.add_argument("--half", action="store_true", help="use FP16 half-precision inference")
|
296 |
+
parser.add_argument("--dnn", action="store_true", help="use OpenCV DNN for ONNX inference")
|
297 |
+
parser.add_argument("--vid-stride", type=int, default=1, help="video frame-rate stride")
|
298 |
+
opt = parser.parse_args()
|
299 |
+
opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1 # expand
|
300 |
+
print_args(vars(opt))
|
301 |
+
return opt
|
302 |
+
|
303 |
+
|
304 |
+
def main(opt):
|
305 |
+
"""Executes YOLOv5 model inference with given options, checking requirements before running the model."""
|
306 |
+
check_requirements(ROOT / "requirements.txt", exclude=("tensorboard", "thop"))
|
307 |
+
run(**vars(opt))
|
308 |
+
|
309 |
+
|
310 |
+
if __name__ == "__main__":
|
311 |
+
opt = parse_opt()
|
312 |
+
main(opt)
|