This codebase implements the system described in the paper:
DPSNet: End-to-end Deep Plane Sweep Stereo
Sunghoon Im, Hae-Gon Jeon, Steve Lin, In So Kweon
In ICLR 2019.
See the paper for more details.
Please contact Sunghoon Im (sunghoonim27@gmail.com) if you have any questions.
Building and using requires the following libraries and programs
Pytorch 0.3.1
CUDA 9.0
python 3.6.4
scipy
argparse
tensorboardX
progressbar2
path.py
The versions match the configuration we have tested on an ubuntu 16.04 system.
Training data preparation requires the following libraries and programs
opencv
imageio
joblib
h5py
lz4
- Download DeMoN data (https://github.com/lmb-freiburg/demon)
- Convert data
[Training data]
bash download_traindata.sh
python ./dataset/preparation/preparedata_train.py
[Test data]
bash download_testdata.sh
python ./dataset/preparation/preparedata_test.py
python train.py ./dataset/train/ --mindepth 0.5 --nlabel 64 --log-output
python test.py ./dataset/test/ --sequence-length 2 --output-print --pretrained-dps ./pretrained/dpsnet.pth.tar