search

YufeiWang777 / LRRU

Official implementation of ``LRRU: Long-short Range Recurrent Updating Networks for Depth Completion'', ICCV 2023.
83 stars 4 forks source link

Detach problems #13

Open jisoo0-0 opened 9 months ago

jisoo0-0 commented 9 months ago

Hi, thank you for your wonderful work.

I have some questions about the Python file"model_dcnv2.py". You have detached some outputs from conv, but I am wondering why. Because if you detach the torch, the model cannot be updated based on that torch.

Thanks.

YufeiWang777 commented 5 months ago

Hi, the updating unit refines the depth map by learning the parameter of the spatial propagetion process. Since it is very difficult to optimize both the depth map and the parameters at the same time, I detach the depth map.