search

YanFangCS / AINET

Other
23 stars 1 forks source link

if no boundary perceiving loss is used #2

Open fanrz opened 2 years ago

fanrz commented 2 years ago

Hi, thanks for your work. if no boundary perceiving loss is used, will the result keep the clear edge and show superpixel clustering in semantic segmentaion task? do you think this method work well in depth estimation task?

wangyxxjtu commented 1 year ago

Hi Fanrz, without the boundary perceiving loss, the edge will also be clear, this could be observed in SCN model. For the second question, it may be also helpful. Since we apply our AINet to the stereo match task, it helps.

fanrz commented 1 year ago

thanks for your reply. I make my idea clearer. I hope to use your method in one depth estimation task, and want to get clear edge as images. In KITTI task, the depth groundtruth is semi-dense, usually no supervision in edge region. So I hope to use superpixel, as some papers have used similar methods And your paper link is wrong, it links to "Superpixel Segmentation with Fully Convolutional Networks".

wangyxxjtu commented 1 year ago

thanks for your reminder, we have corrected the link.