search

AkiraHero / diffindscene

[CVPR2024] DiffInDScene: Diffusion-based High-Quality 3D Indoor Scene Generation
https://akirahero.github.io/diffindscene/
51 stars 1 forks source link

Good Jobs. Can you provide start guide? I encounter some package loss and conflicts when using dev_env.yml #2

Open miyapeng opened 3 months ago

miyapeng commented 3 months ago

Good Jobs. Can you provide start guide? I encounter some package loss and conflicts when using dev_env.yml just like Pip subprocess error: ERROR: Could not find a version that satisfies the requirement diff-gaussian-rasterization==0.0.0 (from versions: none) ERROR: No matching distribution found for diff-gaussian-rasterization==0.0.0 and so on Thank you!

RuihanLu commented 3 months ago

Giving my solution as a reference:

Run conda env update -f dev_env.yml multiple times, commenting out the packages that appear in the error report from dev_env.yml until the command runs successfully.

Then just go ahead and install those commented packages manually. Those packages can be found on GitHub with detailed tutorials on how to install them.

miyapeng commented 3 months ago

Thank you. I solve the problem finally. The paper have not release the checkpoint , have you train?I wander how many GPUs and how much memoey needed?

RuihanLu commented 3 months ago

I haven't started training yet because data processing is slow. I'm still using Blender to remesh the scene to be a watertight mesh. There are about 6,000 scenes. I've run it for 50 hours so far and finished extracting 3000 scenes.

miyapeng commented 3 months ago

That really spends long time! will you release your processed data? Thank you very much

RuihanLu commented 3 months ago

Maybe not. Those files are too large, like 1.2TB, and I don't have enough space on Google Drive.