The code of Ada-MVS for "Deep learning based multi-view stereo matching and 3D scene reconstruction from oblique aerial images (ISPRS) "
paper link:ISPRS_2023
This code is an implementation of Ada-MVS (adaptive multi-view aggregation matching network), which aims to infer the reference depth map from a set of oblique aerial multi-view images with known imaging parameters.
- python 3.7
- pytorch >= 1.3.1
- opencv-python >= 4.1
- numpy >= 1.19
- Download the WHU Oblique MVS dataset. http://gpcv.whu.edu.cn/data.
- Unzip the dataset to the
WHU_OMVS
folder.
WHU_OMVS
├── train
├── test
├── predict
- In
train_whu.py
, setmode
totrain
, setmodel
toadamvs
, setset_name
towhu_omvs
- Set
trainpath
to your training data folderYOUR_PATH/WHU_OMVS/train
, and settestpath
to your validation data folderYOUR_PATH/WHU_OMVS/test
- Set
loadckpt
toNone
- Train the model:
python train_whu.py
- In
train_whu.py
, setmode
totest
, setmodel
toadamvs
, setset_name
towhu_omvs
- set
testpath
to your test data folderYOUR_PATH/WHU_OMVS/test
- set
loadckpt
as your checkpoint file - Test the model:
python train_whu.py
The test outputs are stored in YOUR_PATH/WHU_OMVS/test/depths_whu_omvs/
, including depth map XXX_init.pfm
, probability map XXX_prob.pfm
and the visual results in color
folder.
- In
predict_whu.py
, setmodel
toadamvs
. - set
data_folder
to the test data information folderYOUR_PATH/WHU_OMVS/predict/source
, setoutput_folder
to your save pathOUTPUT_PATH
- set
loadckpt
as your checkpoint file - Run inference:
python predict_whu.py
The inferred results are stored in OUTPUT_PATH
, including depth map XXX_init.pfm
, probability map XXX_prob.pfm
and the visual results in color
folder.
If you find this code useful in your research, please cite:
@article{liu_deep_2023,
author = {Liu, Jin and Gao, Jian and Ji, Shunping and Zeng, Chang and Zhang, Shaoyi and Gong, JianYa},
title = {Deep learning based multi-view stereo matching and {3D} scene reconstruction from oblique aerial images},
volume = {204},
issn = {09242716},
journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
month = oct,
year = {2023},
pages = {42--60},
}
and
@inproceedings{liu_novel_2020,
author = {Liu, Jin and Ji, Shunping},
title = {A Novel Recurrent Encoder-Decoder Structure for Large-Scale Multi-View Stereo Reconstruction From an Open Aerial Dataset},
booktitle = {2020 {IEEE}/{CVF} Conference on Computer Vision and Pattern Recognition ({CVPR})},
month = {June},
year = {2020},
pages = {6049--6058}
}
This project takes the CasMVSNet and REDNet as its backbone, and we thank the authors for providing the source code.