iconvN layer does not have prN+1 as input

[saxenarohan97]

Hi @ZhijianJiang, Thank you for sharing your code.

I had a question about the upconvolutional part of your network. In the paper, it is mentioned that each iconvN layer takes as input upconvN, prN+1 and convN as inputs:

However, in your implementation, you have only used upconvN and convN as the inputs. Is there a reason for this?

[candlecove-nju]

Hi @ZhijianJiang, Thank you for sharing your code. I had a question about the upconvolutional part of your network. In the paper, it is mentioned that each iconvN layer takes as input upconvN, prN+1 and convN as inputs:

However, in your implementation, you have only used upconvN and convN as the inputs. Is there a reason for this?

I think it's a mistake, so I add prx layer to his dispnet. In theory, concat prx to iconv layer can accelarate training rate because prx convey low resolution disparity prediction to next part.