https://github.com/werserk/mayachallenge

Solution of my first DL competition - MayaChallenge. Task is segmentation of Maya buildings on satellite images.
https://github.com/werserk/mayachallenge

deep-learning satellite-images segmentation torch

Last synced: 3 months ago
JSON representation

Solution of my first DL competition - MayaChallenge. Task is segmentation of Maya buildings on satellite images.

• Host: GitHub
• URL: https://github.com/werserk/mayachallenge
• Owner: werserk
• License: mit
• Created: 2021-09-02T13:54:09.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-12-23T14:50:57.000Z (over 2 years ago)
• Last Synced: 2025-01-14T03:13:20.803Z (4 months ago)
• Topics: deep-learning, satellite-images, segmentation, torch
• Language: Jupyter Notebook
• Homepage: https://biasvariancelabs.github.io/maya_challenge/
• Size: 1.76 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# MayaChallenge



## Competition overview

Task is semantic segmentation of Maya building on satellite images.

The dataset consists of tiles derived from of Sentinel-1, Sentinel-2, and ALS (lidar) data, and associated annotation

masks. Each tile measures 240 x 240 meters and has spatial resolution of 10 meters for Sentinel data and 0.5 meters for

ALS data. The Sentinel-1 and Sentinel-2 data for each tile is stored separately in multi-band TIFF files (see data

structure).

| aguadas | platforms | buildings |

| :-----: | :-------: | :-------: |

| ![](.images/tile_44_mask_aguada.png) | ![](.images/tile_44_mask_platform.png) | ![](.images/tile_44_mask_building.png) |

## Methods

I was trying to use different models and different approaches to solve this task.

models:

- Unet

- Linknet

- FPN

- PSPNet

- DeepLabV3 (best)

losses:

- BCE

- Dice

- Focal

- IoU (best)

optimizers:

- AdamW

- SGD

- Adam (best)

## Results

I have reached these metrics and 9th place in the competition.

| Avg. IOU (overall) | Avg. IOU of aguadas | Avg. IOU of platforms| Avg. IOU of buildings |

|---|---|---|---|

| 0.7905 | 0.9718 | 0.6983 | 0.7013 |