Skip to content

Networks

Jingyu Song edited this page Feb 17, 2022 · 5 revisions

MotionSC

    • Backbone of MotionSC, a slightly modified spatio-temporal pyramid network from (2).
    • MotionSC model. The backbone is a STPN (2), and the head is a semantic segmentation head (3).

LMSCNet

    • LMSCNet model, implementation from (3).
    • Single stage version, as it has better performance at full resolution.

SSCNet

    • SSCNet (6) model, with implementation from (3).
    • SSCNet (6) full model, with implementation from (3). Deconvolution is added to the output to create a finer resolution prediction.

JS3C-Net

The JS3C-Net is one of the state-of-the-art algorithms on semantic scene completion. However, due to its complicated dependencies and outdated libraries used, we decide to separate it from this package. We made a forked version of JS3C-Net for the CarlaSC dataset. You can check the readme of that repository on how to run training and testing.

Clone this wiki locally