Add SqueezeSeg for Semantic Segmentation of 3D LiDAR point clouds

[soumik12345]

Short Description

Baseline models for end-to-end semantic segmentation of 3D LiDAR point clouds, specifically (but not limited to) the detection and categorization of instances of interest, such as cars, pedestrians, and cyclists. Also, data-loading utilities for large-scale labeled datasets of 3D LiDAR point clouds such as Semantic-KITTI and nuScenes could be added.

Papers

• SqueezeSeg: Convolutional Neural Nets with Recurrent CRF for Real-Time Road-Object Segmentation from 3D LiDAR Point Cloud
• SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud

Existing Implementations

• TensorFlow implementation

Other Information

• This proposal is inspired by the coursework from Automated and Connected Driving Challenges.
• I am willing to contribute.

[divyashreepathihalli]

@soumik12345 Thanks for opening this feature request. If you are interested in contributing please go ahead and open the PR. This is not a priority for the team at the moment, so we will not be able to guide you with the implementation. But if you do have a PR, we can help review and add it to our code base.

[soumik12345]

Thanks, I will start working on this PR.