binarization augmentation: binarize channels together

[bertsky]

IMO this is incorrect and can create bad results which could then deteriorate training:

sbb_pixelwise_segmentation/utils.py

Lines 95 to 106 in 63fcb96

	def otsu_copy(img):
	img_r=np.zeros(img.shape)
	img1=img[:,:,0]
	img2=img[:,:,1]
	img3=img[:,:,2]
	_, threshold1 = cv2.threshold(img1, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
	_, threshold2 = cv2.threshold(img2, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
	_, threshold3 = cv2.threshold(img3, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
	img_r[:,:,0]=threshold1
	img_r[:,:,1]=threshold1
	img_r[:,:,2]=threshold1
	return img_r

Thresholding must not apply to each color channel separately, but all combined – e.g. by doing grayscale reduction first.