LLR-MVSNet

A ligweight network for low-texture reconstruction

How to Use

Environment

• python 3.8
• pytorch 1.12.1

Training

• Download the preprocessed DTU training data and Depths_raw (both from Original MVSNet), and upzip it as the $MVS_TRANING folder.

• in train.sh, set MVS_TRAINING as your training data path

• Train LLR-MVSNet: scripts/train.sh ##Testing

• Download our pre-processed dataset: DTU's testing set (from Original MVSNet) ,Tanks & Temples and ETH3D benchmark. Each dataset is already organized as follows:

root_directory
├──scan1 (scene_name1)
├──scan2 (scene_name2) 
      ├── images                 
      │   ├── 00000000.jpg       
      │   ├── 00000001.jpg       
      │   └── ...                
      ├── cams                   
      │   ├── 00000000_cam.txt   
      │   ├── 00000001_cam.txt   
      │   └── ...                
      └── pair.txt  

• In scripts/test.sh, set DTU_TESTPATH as $DTU_TESTPATH.
• The DTU_CKPT_FILE is automatically set as your pretrained checkpoint file, you also can download my pretrained model.
• Test on GPU by running scripts/test.sh. The code includes depth map estimation and depth fusion. The outputs are the point clouds in ply format.
• For quantitative evaluation on DTU dataset, download SampleSet and Points. Unzip them and place Points folder in SampleSet/MVS Data/. The structure looks like:

SampleSet
├──MVS Data
      └──Points

In evaluations/dtu/BaseEvalMain_web.m, set dataPath as path to SampleSet/MVS Data/, plyPath as directory that stores the reconstructed point clouds and resultsPath as directory to store the evaluation results. Then run evaluations/dtu/BaseEvalMain_web.m in matlab.

• For quantitative evaluation on Tanks & Temples and ETH3D benchmark, please submit to the website.

Results on DTU

	Acc.	Comp.	Overall.
CasMVSNet	0.325	0.385	0.355
LLR-MVSNet	0.314	0.318	0.316

Results on Tanks and Temples benchmark

Mean	Family	Francis	Horse	Lighthouse	M60	Panther	Playground	Train
60.7	80.09	63.28	53.27	57.74	60.74	57.63	54.93	57.91

Results on ETH3D benchmark

Acknowledgements

Our work is partially baed on these opening source work: MVSNet, MVSNet-pytorch, cascade-stereo, PatchmatchNet，MVSTER.

We appreciate their contributions to the MVS community.