MonocularDepth

This repository is for testing and comparing different approaches to monocular depth estimation.

Requirements

Clone repository with submodules (git clone --recurse-submodules ...).

Install the submodules in editable mode

pip install -e monodepth2 mannequinchallenge
pip install torch torchvision scikit-image h5py

Usage

monodepth2

Use infer_depth(model, img) from infer.py to extract a depth array and a disparity image from an RGB image. The depth values will need to be scaled by a scalar value. For stereo-trained models on KITTI, this value is 5.4.

mannequinchallenge

Retrieve checkpoints by

cd mannequinchallenge
./fetch_checkpoints.sh
cd ..

Use infer_depth(img) from infer.py to extract a depth array and a disparity image from an RGB image.