Awesome

Path Invariance Network in Tensorflow

Tensorflow implementation of one application in Path Invariance Map Networks: 3D Scene Semantic Segmentation. Since we use pointnet2 codebase, some of the code are borrowed from there.

alt tag

Prerequisites

Two GPUs(GTX 1080 or better)
Python 2.7
Tensorflow 0.12.1
SciPy
Pointnet2 Customized TF Operators Please make sure the tf_ops folder is placed in the same directory.

Usage

First, download dataset here. (Point clouds are collected by Pointnet++) The pre-trained models can be downloaded here. (Point clouds models are trained using Pointnet++, and voxel models are trained using 3D-U-Net. The training code for voxel models will be released soon.)

To train models with downloaded dataset:

$ ./cmd

All training commands are included in the cmd file. Testing results are logged during the training stage.

Results

alt tag

Author

Zaiwei Zhang

License

Our code is released under BSD License (see LICENSE file for details).