Are now doing a semantic segmentation task, the goal is to identify extracted from the image, the R, G, B three channel and ground of gray image way are white, the background is black, if R channel into 1 - R (namely is the higher the probability of the closer it gets to the black, the higher the probability of the background of the more close to white), as a 1 - R, G, B three channel input to the network, please will influence on the results and how to, the impact is because what reason, thank you
The idea is to reduce the segmentation accuracy, but want to detail again and said don't understand, R G B three-channel grayscale although there are nuances, but basically 1 - R should be roughly a channel offset, finally left a channel combining two basic offset some of the subtle information, but this one channel information network can also study well,