DI-MVS (ICASSP 2024)

Installation

git clone https://github.com/JianfeiJ/DI-MVS.git
cd DI-MVS

Enviorment

conda create -n di-mvs python=3.8
conda activate di-mvs
pip install -r requirements.txt

Datasets

DTU

• Download pre-processed datasets for test: dtu.

  - dtu/
  - scan1 (scene_name1)
  - scan2 (scene_name2)
    - images
      - 00000000.jpg
      - 00000001.jpg
      - ...
    - cams_1
      - 00000000_cam.txt
      - 00000001_cam.txt
      - ...
    - pair.txt
  

• Download pre-processed datasets for training: dtu_training, Depth_raw.

  - dtu_training/
  - Cameras
  - Depths
  - Depths_raw
  - Rectified
  

Tanks & Temples

Download Tanks & Temples test dataset.

- tanksandtemples_1/
 - advanced
   - ...
   - Temple
     - cams
     - images
     - pair.txt
     - Temple.log
 - intermediate
   - ...
   - Train
     - cams
     - cams_train
     - images
     - pair.txt
     - Train.log

Reproducing Results (single RTX 3090)

Download pretrained model on DTU and BlendedMVS, and remove them to the checkppints folder.

Evaluation on DTU

sh test_dtu.sh

For quantitative evaluation on DTU dataset, download SampleSet and Points. Unzip them and place Points folder in SampleSet/MVS Data/.

- SampleSet
  - MVS Data
    - ObsMask
    - Points

Evaluation on Tanks & Temples

sh test_tnt.sh

Result on DTU

Methods	Acc. (mm)	Comp. (mm)	Overall (mm)	Time (s)
DI-MVS-lite	0.305	0.305	0.305	0.11
DI-MVS	0.312	0.278	0.295	0.30

Result on Tanks & Temples benchmark.

Intermediate	Advanced
62.94	40.92

Traning

sh train.sh

Acknowledgements

This repository is partly based on Effi-MVS, TransMVSNet, IterMVS, UniMVSNet and MVSTER. Thanks for their excellent works!