Small-3D-Semantic-Segmentation-model

Implement a small 3D Semantic Segmentation model using Sparse Convolutions and overfit (train and eval on the same data) on the given dataset.

Environment: ubuntu20.04,python3.8

INSTALL

Clone the repository and the 'data.npy' is located in ~/dataset/raw/ by default

git clone https://OrangeSodahub.com/Open3D_assignment.git

Then install additional required packages

cd ~/
python install -r requirements

Usage

Generate Datasets

First generate train dataset and test dataset from given dataset data.npy. Default data file path is given.

cd ~/
python tools/generate_dataset.py

To specify the path to config file, use config --config. And it also support the visualization of the raw dataset via open3d, if you want just add --visible

python tools/generate_dataset.py --config /path/to/config/file --visible True

Train

Train using the given data file.

python train.py

Test

Test on the given data file using trained model.

python test.py

OrangeSodahub / Small-3D-Semantic-Segmentation-model