theSoenke / deep-stereo-matching

PyTorch implementation of "Efficient Deep Learning for Stereo Matching"

deep-stereo-matching

Preprocess data

1. Matlab or Octave needs to be installed
2. Replace path to dataset in preprocess/preprocess.m
3. cd preprocess
4. octave preprocess.m

Train

python3 train.py --data /data/kitti2015/training

Test

python3 eval.py --data /data/kitti2015/training --checkpoint pretrained pretrained/win37_dep9.pkl

Inference

python3 inference --data /data/kitti2015/testing --img_num 0

Results

Iterations: 40000, Batch size: 128, Depth: 9, Kernel size: 5 x 5