Please download KITTI dataset and organize the data as follows:
#MonoDTR_ROOT
|data/
|KITTI/
|object/
|training/
|calib/
|image_2/
|label_2/
|velodyne/
|testing/
|calib/
|image_2/
You can modify the path in config/config.py (for train / val split), and then run the preparation script:
cd #MonoDTR_ROOT
./launchers/det_precompute.sh config/config.py train # precompute image database and anchors mean/std
python scripts/depth_gt_compute.py --config=config/config.py # precompute depth gt for training