This is a PyTorch(0.4.1) implementation of Deep Ordinal Regression Network for Monocular Depth Estimation. At present, we can provide train script in NYU Depth V2 dataset and Kitti Dataset!

Note: we modify the ordinal layer using matrix operation, making trianing faster.

• DORN model in nyu and kitti
• Training DORN on nyu and kitti datasets
• Results evaluation on nyu test set
• the script to generate nyu and kitti dataset.
• Calculate alpha and beta in nyu dataset and kitti dataset
• Realize the ordinal loss in paper

DORN need to use all the Images (about 120k) in the dataset, but if you just want to test the code, you can use the nyu_depth_v2_labeled.mat and turn it to a h5 file. The convert script is 'create_nyu_h5.py' and you need to change the file paths to yours.

• Modify create_nyu_h5.py with your path and run the script.

python create_nyu_h5.py

The kitti dataset contains 23488 images from 32 scenes for training and 697 images from the remaining 29 scenes for testing.

• Raw dataset (about 175 GB) can be downloaded by running:

wget -i kitti_archives_to_download.txt -P ~/kitti-raw-data/

• Unzip the compressed files:

cd ~/kitti-raw-data
find . -name '*.zip' -exec unzip {} \;

• Run the script to generate the kitti_ground_truth

python gen_kitti_dataset.py