JonasSchult / Mask3D

Mask3D predicts accurate 3D semantic instances achieving state-of-the-art on ScanNet, ScanNet200, S3DIS and STPLS3D.
MIT License
545 stars 110 forks source link

Demo results #82

Open irineu93 opened 1 year ago

irineu93 commented 1 year ago

Hello,

Thank you very much for your work on the paper and repository.

I was wondering if, in the case of the online demo, you used the same checkpoints as provided here, in the home page of the GitHub project. More specifically I was thinking of the Scannet200 benchmark checkpoint. The reason why I'm asking is because when I tested locally with a personal 3D scene, the results were a bit different compared to the online demo (missing / other new predictions).

So, I was wondering if the cause is a different checkpoint or maybe just other configuration parameters (I've played around a bit with them as well).

Moreover, the online demo doesn't work anymore. More specifically, like some other people mentioned in other issues, after the scene is uploaded and preprocessed, the generated links for visualization and predictions download are not valid (because the resources can't be found on the server).

Thank you.

bh-cai commented 1 year ago

Dear predecessors, I am new to this code, may I know how to make the datasets by using our own data, and what format of the dataset we make by ourselves? Thank you very much !

bh-cai commented 1 year ago

Dear predecessors, I am new to this code, how can we make the datasets by using our own data, and what format of the dataset we make by ourselves? Thank you very much!

LYX0501 commented 5 months ago

The same question about demo results. Have you solved it? @irineu93