MRD-Depth

This repository provides an official source code for:

Multi-Resolution Distillation for Self-Supervised Monocular Depth Estimation
Pattern Recognition Letters, 2023

The code base is Monodepth2 (Godard et al. 2019).

⏳ Preparation for Training and Evaluation

Please follow the several steps for training and evaluation.

1> Download KITTI raw dataset and unzip it.

wget -i splits/kitti_archives_to_download.txt -P kitti_data/
cd kitti_data
unzip "*.zip"
cd ..

2> Convert the image format to png.

find kitti_data/ -name '*.png' | parallel 'convert -quality 92 -sampling-factor 2x2,1x1,1x1 {.}.png {.}.jpg && rm {}'

3> Download the baseline model (Zhou et al. 2021) and locate them in ./checkpoints/diffnet folder.

4> Extract ground-truth labels for evaluation

python export_gt_depth.py --data_path kitti_data --split eigen

Evaluation

Evaluate our model:

You can evaluate the trained model as follows:

python evaluate_depth.py 
    --data_path [KITTI_DATA_ROOT]
    --eval_mono

Pre-trained Weights

Comming Soon.