search

alexklwong / calibrated-backprojection-network

PyTorch Implementation of Unsupervised Depth Completion with Calibrated Backprojection Layers (ORAL, ICCV 2021)
Other
117 stars 24 forks source link

Why quality is so bad? #25

Open denred0 opened 1 year ago

denred0 commented 1 year ago

Hi!

Thanks for your great work! I have tested your repository and got bad results. I tested with bash/void/run_kbnet_void150.sh

There is the image 0000000043

Depth map: grayscale

Sparse point cloud: screencast-bpconcjcammlapcogcnnelfmaeghhagj-2023 04 21-17_50_16

And resulting dense point cloud: screencast-bpconcjcammlapcogcnnelfmaeghhagj-2023 04 21-17_53_47

Looks much worse than in your README. Is it normal or I don't know about additional postprocessing? Thanks.

alexklwong commented 1 year ago

Hi, looks like there are some issues with torch version and CUDA affecting training and inference. Previously, it was done with GTX family. We are now retraining on RTX with torch 1.10 and will upload the weights.