How to implement ROI support in detect.py

[sergey99si]

How to implement ROI support in detect.py:
Add a new argument for ROI.
In the detect.py file, you need to add an argument to specify the coordinates of the ROI. This can be done using the argparse library.

Filter the detected objects by ROI.
After getting the results from the YOLO model, add a check to see if the objects' frame coordinates fall within the specified region.

add detect.py parser.add_argument('--roi', type=str, default=None, help='Region of interest (ROI) in the format "x_min,y_min,x_max,y_max"')

add detect.py def parse_roi(roi_str):
if roi_str:
coords = list(map(int, roi_str.split(',')))
if len(coords) == 4:
return coords # [x_min, y_min, x_max, y_max]
else:
raise ValueError('ROI should have 4 integers: x_min,y_min,x_max,y_max')
return None

roi = parse_roi(opt.roi)

for i, det in enumerate(pred): # Перебор результатов
if len(det):
for *xyxy, conf, cls in reversed(det):
x_min, y_min, x_max, y_max = map(int, xyxy)
if roi: # Фильтруем только объекты в ROI
if x_min < roi[0] or y_min < roi[1] or x_max > roi[2] or y_max > roi[3]:
continue # Пропускаем объекты вне ROI

Arguments for ROI in the new script:
--roi x_min,y_min,x_max,y_max: sets the region of interest where objects will be saved.
--save-crop: saves cut images of found objects.
--save-txt: saves a text file with the results.

example:python detect.py --source image.jpg --roi 100,100,500,500 --save-crop