xuanjiawang / Occ-SDF-Hybrid

This is the officially implementation of ICCV 2023 paper " Learning A Room with the Occ-SDF Hybrid: Signed Distance Function Mingled with Occupancy Aids Scene Representation"

Learning A Room with the Occ-SDF Hybrid: Signed Distance Function Mingled with Occupancy Aids Scene Representation

Xiaoyang Lyu · Peng Dai · Zizhang Li · Dongyu Yan · Yi Lin · Yifan (Evan) Peng · Xiaojuan Qi

ICCV 2023

Paper | Project Page

TL; DR: Implicit neural rendering has made impressive strides in the surface reconstruction of large-scale scenes. However, applyding previous method to reconstruct a room-level scene from images may miss structures in low-intensity areas and/or small, thin objects. We have conducted experimetns on three datasets to identify limitations of the original color rendering loss and priors-embedded SDF scene representation.

TODO

• Create the project page
• Create the README file
• Opensource all the training code
• Opensource all the models and meshes
• Replica data preparation scripts
• Real data preparation scripts
• MonoUNISURF and MonoVolSDF

Setup

Installation

Clone the repository and create an anaconda environment called occ-sdf using

git clone [e-mail]:shawLyu/Occ-SDF-Hybrid.git
cd Occ-SDF-Hybrid

conda env create -f env.yaml

The hash encoder will be compiled on the fly when running the code.

Dataset

This project mainly inherits from the MonoSDF, thus the dataset can be downloaded like Monosdf, through

bash scripts/download_dataset.sh

Training

Run the following command to train OCC-SDF:

cd ./code
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --nproc_per_node 1 --nnodes=1 --node_rank=0 training/exp_runner.py --conf CONFIG --scan_id SCAN_ID 

where CONFIG is the config file in code/confs, and SCAN_ID is the id of the scene to reconstruct.

If you want to train the network on the single machine with different GPU, you can add --master_port <port_number>

We provide example commands for training Replica, ScanNet, and Tanks and Temples dataset as follows:

# Replica (room0)
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_port 25585 --nproc_per_node 1 --nnodes=1 --node_rank=0 training/exp_runner.py --conf confs/replica_mlp_hybrid.conf --scan_id 1

# Scannet (scene_0050_00)
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_port 25585 --nproc_per_node 1 --nnodes=1 --node_rank=0 training/exp_runner.py --conf confs/scannet_mlp_hybrid.conf --scan_id 1

# Tanks and Temples (MLP)
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_port 25585 --nproc_per_node 1 --nnodes=1 --node_rank=0 training/exp_runner.py --conf confs/tnt_mlp_hybrid.conf --scan_id 1

# Tanks and Temples (Grids)
CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_port 25585 --nproc_per_node 1 --nnodes=1 --node_rank=0 training/exp_runner.py --conf confs/tnt_grids_hybrid.conf --scan_id 1

Evaluations

Replica

Evaluate on one scene (take scan 1 room0 for example)

cd replica_eval
python evaluate_single_scene.py --input_mesh replica_scan1_mesh.ply --scan_id 1 --output_dir replica_scan1

Evaluate all Replica Scenes:

cd replica_eval
python evaluate_avg.py

please check the script for more details.

ScanNet

cd scannet_eval
python evaluate_avg.py

please check the script for more details.

Tanks and Temples

Please follow the MonoSDF for more details.

Acknowledgements

This project is built upon MonoSDF. We thank all the authors for their great work and repos.

Citation

If you find our code or paper useful, please cite

@inproceedings{Lyu2023occsdf,
  title={Learning A Room with the Occ-SDF Hybrid: Signed Distance Function Mingled with Occupancy Aids Scene Representation},
  author={Xiaoyang Lyu, Peng Dai, Zizhang Li, Dongyu Yan, Yi Lin, Yifan Peng, Xiaojuan Qi},
  booktitle={Proceedings of the IEEE/CVF international conference on computer vision},
  year={2023}
}