Small object detection - what is the minimum bounding box size?

[DSA101]

What is the smallest bounding box size for Yolo5 training? I'll need to detect medium and small objects (think cars at distance or ships at distance) that can be just a couple of pixels in size. It'll be just one class of objects that I'll need to detect.

[glenn-jocher]

@DSA101 default label candidate criteria includes >2 pixels. This can be manually modified here:

yolov5/utils/augmentations.py

Lines 272 to 277 in 53349da

	def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
	# Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
	w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
	w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
	ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio
	return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates