Training w/o segments and evaluating w segments

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 w/o segments and evaluating w segments #149

Open 0nandon opened 8 months ago

0nandon commented 8 months ago

Hi Jonas. Thanks for ur great work.

If I train the model w/o segments and evaluate it with segments, how would the performance be compared to the official performance reported in the paper?

In the paper, model is trained with segments and evaluated with segments, right?

Would be the performance roughly same?