This Week in 3D Imaging

|
Antonina Pevzner
|
0 comments

Research Highlights 

  • In order to effectively capture the global multi-scale information and learn sequence representations of the input volume, researchers from NVIDIA introduced a novel architecture known as UNEt TRansformers (UNETR). It uses a transformer as the encoder and adheres to the successful "U-shaped" network design for the encoder and decoder. 

  • Both numerically and qualitatively, researchers from MagicLink greatly surpassed state-of-the-art baselines (deep multiview stereo followed by standard TSDF fusion). Since no other work has attempted the challenge using only RGB input, we compare our 3D semantic segmentation against earlier techniques that employ a depth sensor. 

 

Open Source Highlights 

  • The Sapling team revealed that the Sapling client would now be open-sourced. With Sapling's integrated Git support, you can now clone any of your current repositories and test out its different capabilities. This marks the beginning of a more involved endeavor to make the complete Sapling system accessible to everyone.