MonocularDepthEstimation

KO | EN

gitlite — search

#python #typescript #golang #go #rust #deep-learning #linux #nodejs #awesome #pytorch #security #ios #mcp #openai

MonocularDepthEstimation

(★ 14)

In this repository, monocular depth estimation is implemented using MiDaS v2.1 small model for Android mobile devices.

MonocularDepthEstimation Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

[![DOI](https://zenodo.org/badge/580930775.svg)](https://doi.org/10.5281/zenodo.18993520)

# Monocular Depth Estimation With MiDaS on Android Mobile Devices
This project is developed for the [CMP721 - Computational Photography](http://www.ahmetselman.com/cmp721/index.html) (Fall 2022) PhD course given by [Ahmet Selman Bozkır](http://www.ahmetselman.com) from [Computer Engineering Department](http://cs.hacettepe.edu.tr) at [Hacettepe University](http://hacettepe.edu.tr).

## Introduction
In this project, monocular depth estimation is implemented using pre-trained MiDaS deep learning model. Tensorflow Lite is used as a backend inference engine in the project. The engine uses the [MiDaS](https://github.com/isl-org/MiDaS) v2.1 small as an inference model. The model can be downloaded from model's [Tensorflow Hub page](https://tfhub.dev/intel/midas/v2_1_small/1).

## Performance
The model performance mostly depends on device specifications that run the application. The model tested on two devices detailed in table below:

| **Device**               | **OS**     | **Inference Time (ms)** | **Frame Per Second** |
|--------------------------|------------|-------------------------|----------------------|
| Xiaomi Redmi Note 10 Pro | Android 13 | 90-150                  | 4                    |
| Xiaomi Redmi Note 8 Pro  | Android 11 | 70-80                   | 6                    |
| Lenovo K6 Note           | Android 7  | 500-700                 | 1                    |
| Galaxy A02S              | Android 12 | Failed                  | Failed               |
| Galaxy J4                | Android 10 | 2400-2800               | 1                    |

## Screenshots

### Lenovo K6 Note
<img src="media/k6note_depth_est_1_gs.jpg" width="400"> <img src="media/k6note_depth_est_1_hm.jpg" width="400">
<img src="media/k6note_depth_est_2_gs.jpg" width="400"> <img src="media/k6note_depth_est_2_hm.jpg" width="400">

### Xiaomi Redmi Note 8 Pro
<img src="media/note8pro_depth_est_prait_gs.jpg" width="400"> <img src="media/note8pro_depth_est_prait_hm.jpg" width="400">
<img src="media/note8pro_depth_est_lscape_gs.jpg" width="400"> 
<img src="media/note8pro_depth_est_lscape_hm.jpg" width="400">

### Xiaomi Redmi Note 10 Pro
<img src="media/note10pro_depth_est_prait_gs_1.jpg" width="400"> <img src="media/note10pro_depth_est_prait_hm.jpg" width="400">
<img src="media/note10pro_depth_est_prait_gs_2.jpg" width="400"> <img src="media/note10pro_depth_est_prait_gs_3.jpg" width="400">
<img src="media/note10pro_depth_est_lscape_gs.jpg" width="400"> 
<img src="media/note10pro_depth_est_lscape_hm.jpg" width="400">

### Galaxy J4
<img src="media/galaxyj4_depth_est_prait_gs.jpg" width="400"> <img src="media/galaxyj4_depth_est_prait_hm.jpg" width="400">
<img src="media/galaxyj4_depth_est_lscape_gs.jpg" width="400"> 
<img src="media/galaxyj4_depth_est_lscape_hm.jpg" width="400">

MonocularDepthEstimation

MonocularDepthEstimation Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

MonocularDepthEstimation

MonocularDepthEstimation Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README