JingZhang617 / UCNet

UC-Net: Uncertainty Inspired RGB-D Saliency Detectionvia Conditional Variational Autoencoders, CVPR2020
179 stars 25 forks source link

Can UCNet test RGB image without depthinformation #11

Open HAOCHENYE opened 3 years ago

JingZhang617 commented 3 years ago

the current code is for rgbd saliency. you can remove the depth branch and train with rgd saliency training dataset, and then test with the rgb image.

On Mon, 25 Jan 2021 at 22:45, Mashiro notifications@github.com wrote:

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/JingZhang617/UCNet/issues/11, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6B4F6QGVAJMNGMTBK3QVTS3VKVJANCNFSM4WRT3RYQ .

-- Jing Zhang Ph.D. Student College of Engineering and Computer Science, Australian National University. Email: zjnwpu@gmail.com

zeroRains commented 2 years ago

Hi, I found that the generator(model) have 4 modules that have used the depth map. I do not know which is the depth branch.

I have tried to change the input channel of the 4 modules to make the depth map be useless. It can train before 10 epochs, then it gets the nan params. It confuses me very much. Please tell me more detail. Thank you.