Authors: Ganlin Zhang, Deheng Zhang, Feichi Lu, Anqi Li
Our project is mainly about reducing the memory usage of nice-slam. Our main contributions includes:
-
Reducing the memory usage by smartly sample the points for rendering.
-
Design and implement sparse version of map interpolation.
-
Design and implement the sparse feature representation (voxel hashing) and integrate it into the nice-slam pipeline.
First you have to make sure that you have all dependencies in place. The simplest way to do so, is to use Anaconda.
You can create an anaconda environment called nice-slam. For linux, you need to install libopenexr-dev before creating the environment.
sudo apt-get install libopenexr-dev
conda env create -f spare_nice_slam.yaml
conda activate nice-slam
Or you can use
sudo apt-get install libopenexr-dev
conda env create -f ./nice-slam-1.0-alpha/environment.yaml
conda activate nice-slam
pip install einops
pip install torch torchvision
Firstly, go to the nice-slam-1.0-alpha folder and run following command:
python -W ignore run.py configs/Demo/demo.yaml
Then, run the following command to visualize.
python visualizer.py configs/Demo/demo.yaml
NOTE: This is for demonstration only, its configuration/performance may be different from our paper.
Alternatively, you can use MeshLab to visualize the mesh output/Demo/mesh/final_mesh.ply .
To evaluate the reconstruction error, first download the ground truth Replica meshes where unseen region have been culled.
bash scripts/download_cull_replica_mesh.sh
Then run the command below. The 2D metric requires rendering of 1000 depth images, which will take some time. Use -2d to enable 2D metric. Use -3d to enable 3D metric.
# assign any output_folder and gt mesh you like, here is just an example
python src/tools/eval_recon.py --rec_mesh output/Replica/room0/mesh/final_mesh_eval_rec.ply --gt_mesh cull_replica_mesh/room0.ply -2d -3d
- We express gratitude to NICE-SLAM, we benefit a lot from both their papers and codes.
- Thanks to Songyou Peng. He has provided many insightful guidance about how to integrate voxel hashing into the nice-slam framework. We would like to express our sincere appreciation to Zihan Zhu, who has provided many intelligent suggestions on this project.