PNXD / FFL-3DOG

Free-form Description-guided 3D Visual Graph Networks for Object Grounding in Point Cloud

FFL-3DOG

Free-form Description-guided 3D Visual Graph Networks for Object Grounding in Point Cloud

We visualize how the 3D grounding performs after VoteNet, nodes pruning and final result.

Setup

The code is now compatiable with PyTorch 1.6.

conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch

Then please run the following command to compile the CUDA module for the PointNet++ backbone.

cd lib/pointnet2

python setup.py install

Data preparation

1. Download the preprocessed GLoVE embeddings and put them under data/.
2. Download the ScanRefer dataset and unzip it under data/.
3. Download the ScanNetV2 dataset and put scans/ under data/scannet/scans/.
4. Running the following command to preprocess ScanNet data.

cd data/scannet/

python batch_load_scannet_data.py

Training

You can train our model by running the scripts

python scripts/train.py --use_color --use_normal --use_pretrained

Thanks to ScanRefer.