phytooracle/3d_semantic_segmentation repository overview

conda create --name open3d-ml python=3.6.9
conda activate open3d-ml
pip install --upgrade pip
pip install open3d==0.11.2
git clone https://github.com/intel-isl/Open3D-ML.git
cd Open3D-ML

pip install -r requirements-tensorflow.txt
# Test it...
python -c "import open3d.ml.tf as ml3d"

cd ~/wherever_you_want_repo_to_live
git clone https://github.com/phytooracle/3d_semantic_segmentation
cd 3d_semantic_segmentation

pip install python-dotenv 

# Edit .env so it has the correct project_dir
cp sample.env .env
vim .env

conda activate open3d-ml
cd directory_with_uncompressed_supervisely_data_directory
python ~/work/repos/3d_semantic_segmentation/src/data/ConvertSuperviselyToGeneric.py
python /home/equant/work/repos/3d_semantic_segmentation/src/data/ConvertSuperviselyToGeneric.py -i full_auto_labeled -o full_auto_labeled-generic

conda activate open3d-ml
cd src/models/
python train_pred_vis.py

cd train_log
cd 00028_RandLANet_SemanticKITTI_tf
tensorboard --logdir=.
firefox http://localhost:6006/

iget -PKVT /iplant/home/travis_simmons/labeled_pcds.tar.gz
tar -zxvf labeled_pcds.tar.gz

singularity build 3d_semantic_training.simg docker://phytooracle/3d_semantic_segmentation
singularity run 3d_semantic_training.simg {directory containing your bin / label files} -m {directory containing your model file} -outdir {where you want the train_pred_vis outputs to land}

git clone https://github.com/phytooracle/3d_semantic_segmentation.git
cd 3d_semantic_segmentation
singularity build 3d_semantic_training.simg .
singularity run 3d_semantic_training.simg {directory containing your bin / label files} -m {directory containing your model file} -outdir {where you want the train_pred_vis outputs to land}

rm 3d_semantic_training.simg
make changes to repo (don't commit)
cd 3d_semantic_segmentation
singularity build 3d_semantic_training.simg .
singularity run 3d_semantic_training.simg {directory containing your bin / label files} -m {directory containing your model file} -o {where you want the train_pred_vis outputs to land}