Submitted anonymously.

Submission data

Full nameSemantic Dense Reconstruction
Descriptionwe use our SLAM system to refine the consistent frame segmentation and feed it to a 2D-3D joint segmentation network based on MinkowskiNet.
Input Data TypesUses Color,Uses Geometry        Uses 2D
Programming language(s)C++&Python with CUDA10.1
HardwareCore AMD-2600X GeFore 2080Ti, 80G RAM
Submission creation date4 Sep, 2021
Last edited4 Sep, 2021

3D semantic label results

Infoavg ioubathtubbedbookshelfcabinetchaircountercurtaindeskdoorfloorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwallwindow
0.6280.8000.6250.7190.5450.8060.4450.5970.4480.5190.9380.4810.3280.4890.4990.6570.7590.5920.8810.7970.634