---
title: (Docs) MiDAS
tags: Documentation
---

<!-- {%hackmd theme-dark %}
{%hackmd sMV2zv-CTsuIqnpb0hZLmA %} -->

# [Documentation] MiDAS
Code can be downloaded here 👇
[Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer](https://github.com/isl-org/MiDaS)

## ⚙️ Environment Setup
1. Use Anaconda to create environment.

conda create --name [ENVNAME] --file=environment.yaml

2. Download any model. I used [dpt_swin2_large_384](https://github.com/isl-org/MiDaS/releases/download/v3_1/dpt_swin2_large_384.pt).
3. Take test image and store in ```input``` path
4. Test.

"""
To infer video file
:param model_type: Name of the model (no need to put the path, just name)
:param input_path: FOLDER of input files
:param output_path: FOLDER of output
:return: .pfm (depth map file) and .png
"""
python run.py \
    --model_type dpt_swin2_large_384 \
    --input_path input/ \
    --output_path output/
    
"""
To infer realtime
"""
python run.py --model_type dpt_swin2_large_384 --side

5. **[Optional]** Error handling.

# No need to run this if you have no error
# ERROR 1: 
# FileNotFoundError: [Errno 2] No such file or directory: './externals/Next_ViT/classification/nextvit.py'
./install_next_vit.sh

# ERROR 2:
# ...
pip install openvino

6. Install ```pypfm``` package to read ```.pfm``` file

pip install pypfm
# pypfm documentation
# https://pypi.org/project/pypfm/

7. Python script to read ```.pfm``` file

from pypfm import PFMLoader

path = "output/photo_1-dpt_swin2_large_384.pfm"

loader = PFMLoader(color=False, compress=False)
pfm = loader.load_pfm(path)
print(pfm)

8. Downgrade Pytorch version to match current CUDA (10.2)

conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=10.2 -c pytorch

## Score
Version | Model | FPS | Image
--- | --- | --- | ---
3.1 | dpt_swin2_large_384 | 7.9 ± 0.2 | 
3.1 | dpt_swin2_tiny_256 | 30.0 ± 3.0 | 
3.0 | dpt_hybrid_384 | 13.5 ± 1.0 | 
3.0 | dpt_large_384 | 7.85 ± 0.1 |