search

p-ranav / merged_depth

Monocular Depth Estimation - Weighted-average prediction from multiple pre-trained depth estimation models
MIT License
48 stars 3 forks source link
academic average average-calculator camera deep-learning depth-estimation depth-prediction image-processing images machine-learning mit-license neural-networks open-source prediction python3 pytorch torch torchvision youtube

readme

merged_depth runs (1) AdaBins, (2) DiverseDepth, (3) MiDaS, (4) SGDepth, and (5) Monodepth2, and calculates a weighted-average per-pixel absolute depth estimation.

Quick Start

First, download the pretrained models using the download_models script.

Next, run the infer script - this will run on all images in test/input and save the results to test/output. 

python3 -m pip install -r requirements.txt
python3 -m merged_depth.utils.download_models
python3 -m merged_depth.infer

If you're using Anaconda3, the following has been tested to work (in Windows):

conda create --name merged_depth python=3.6
conda activate merged_depth
conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=11.1 -c pytorch -c conda-forge
python3 -m pip install -r requirements.txt
python3 -m merged_depth.utils.download_models
python3 -m merged_depth.infer

The results include (1) a _depth.npy file that you can load (see load_and_display_depth.py), (2) a _stacked.png file that shows the original and colorized depth images.

To run the predictor on a single input, use infer_single.py

python3 -m merged_depth.infer_single ~/foo/bar/test.png

Sample Output

The output depth is absolute depth in meters. The colorizer range is [0, 20].