search

JonasSchult / Mask3D

Mask3D predicts accurate 3D semantic instances achieving state-of-the-art on ScanNet, ScanNet200, S3DIS and STPLS3D.
MIT License
525 stars 103 forks source link

Training setup for val.ckpt and test.ckpt generation on ScanNet v2 #156

Open jkini opened 6 months ago

jkini commented 6 months ago

Thanks for your valuable contribution to point cloud research.

I have a question regarding the training setup for the generation of val.ckpt and test.ckpt checkpoints on ScanNet v2. Specifically, I am interested in understanding the differences in the training setup.

Could you provide some details on the distinctions as guided by the configuration file? Also, please clarify whether only the training set data, comprising 1201 scenes, is utilized for training both val.ckpt and test.ckpt checkpoints?