Monocular depth estimation with Self-supervised method - Viettel Digital 2022

1. Overview

• Problem: Depth Estimation

• Method: Monocular depth estimation with self-supervised, based on Monodepth2 and HRDepth. Our proposal methods are detailed in folder document

• Dataset: KITTI

2. Set up

We ran our experiments with PyTorch 1.10.1, CUDA 11.1, Python 3.6.6 and Ubuntu 18.04

KITTI training data

You can download the entire KITTI_raw dataset by running:

wget -i splits/kitti_archives_to_download.txt -P kitti_data/

Then unzip with

cd kitti_data
unzip "*.zip"
cd ..

Warning: it weighs about 175GB, so make sure you have enough space to unzip too!

3. How to run

We have two versions corresponding to VDT_Phase1 and VDT_Phase2.

To run VDT_Phase1

CUDA_VISIBLE_DEVICES=0 python train.py --model_name densenet-hr-depth --split eigen_zhou --backbone densenet --depth_decoder hr-depth --png

To run VDT_Phase2

CUDA_VISIBLE_DEVICES=0 python train_v2.py

📊 KITTI evaluation

To prepare the ground truth depth maps run:

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

...assuming that you have placed the KITTI dataset in the default location of ./kitti_data/.

The following example command evaluates the epoch 19 weights of a model named densenet:

python evaluate_depth.py --load_weights_folder ./densenet/models/weights_19/ --eval_mono --backbone densenet --depth_decoder hr-depth