---
title: Finding the Street-Level Images You Need in Mapillary
tags:  #mapillary #foss4g  
author: [bannyaaa](https://image.docswell.com/user/bannyaaa)
site: [Docswell](https://www.docswell.com/)
thumbnail: https://bcdn.docswell.com/page/KJ4WKD8371.jpg?width=480
description: 2026-09-02  FOSS4G Hiroshima 2026 https://2026.foss4g.org/
published: September 04, 26
canonical: https://image.docswell.com/s/bannyaaa/ZPRJ71-20260902-foss4g-hiroshima
---
# Page. 1

![Page Image](https://bcdn.docswell.com/page/KJ4WKD8371.jpg)

Finding the Street-Level
Images You Need in Mapillary
FOSS4G Hiroshima 2026 LT
Hironori Banno (bannyaaa)


# Page. 2

![Page Image](https://bcdn.docswell.com/page/LE1YXZ2Z7G.jpg)

What is Mapillary?
An open data platform for crowdsourced street-level images
Mapillary Platform
Upload
Download
Open data (CC BY-SA)


# Page. 3

![Page Image](https://bcdn.docswell.com/page/GEWGN9R6J2.jpg)

©︎
©︎
©︎
©︎
©︎
©︎
Mapillary Has Diverse Images
Different subjects and capture conditions
ChenJJ (CC BY-SA)
musashino5 (CC BY-SA)
Ca73 (CC BY-SA)
yasunari (CC BY-SA)
snhoybekcxnb (CC BY-SA)
mitz (CC BY-SA)
We need to select images that fit our purpose


# Page. 4

![Page Image](https://bcdn.docswell.com/page/47ZL39RRJ3.jpg)

Demo Scenario
Find images from central Hiroshima like the example below:
Sampling area
Target image example
Bright
Pedestrian / driver
viewpoint
Non-panoramic
Sharp
@lvl5 (CC BY-SA)
Area data: MLIT National Land Numerical Information (Administrative Areas) modified (CC BY 4.0)
Map data: @OpenStreetMap contributors


# Page. 5

![Page Image](https://bcdn.docswell.com/page/YJ6WXKR1JV.jpg)

Three-Step Filtering Pipeline
Metadata
Low
Non-AI Image
Analysis
AI-based Image
Analysis
Brightness / Blur
Semantic Segmentation
Computational cost
Start with lower-cost steps
High


# Page. 6

![Page Image](https://bcdn.docswell.com/page/GJ5MXP8LJ4.jpg)

©︎
1. Filtering with Metadata
Every Mapillary image has metadata
Image
Metadata
image id:
2157221781753991
captured at:
Feb 7, 2026, 2:11 PM
is panoramic:
False
geometry:
Point (132.47, 34.40)
compass angle: 5.47
maheshika (CC BY-SA)
quality score:
0.69
creator id:
102993915277414
…
…


# Page. 7

![Page Image](https://bcdn.docswell.com/page/9E29X6K37R.jpg)

©︎
©︎
©︎
1. Filtering with Metadata
Selected
Not selected
Daytime
Nighttime
created at:
@drivephotograph (CC BY-SA)
alt9800 (CC BY-SA)
Panoramic
Non-panoramic
is panoramic:
okadatsuneo (CC BY-SA)
potaro67v (CC BY-SA)
Filtering before download reduces later processing


# Page. 8

![Page Image](https://bcdn.docswell.com/page/D7Y489N8EM.jpg)

©︎
©︎
©︎
©︎
2. Filtering with Brightness and Blur
Selected
Bright
Not selected
Slightly dark
Brightness:
okadatsuneo (CC BY-SA)
drivephotograph (CC BY-SA)
Blurred
Sharp
Blur:
hiroshi (CC BY-SA)
yasunari (CC BY-SA)
※Brightness: Mean RGB, Blur: Laplacian variance, Using OpenCV
Simple and fast checks for basic image quality


# Page. 9

![Page Image](https://bcdn.docswell.com/page/VENYQLX9J8.jpg)

©︎
3. Filtering with Semantic Segmentation
Input
lvl5 (CC BY-SA)
Output
Road: 32%, Building: 30%, Vegetation: 15%, …
※Using Mask2Former (facebook/mask2former-swin-tiny-cityscapes-semantic)


# Page. 10

![Page Image](https://bcdn.docswell.com/page/Y79PD483E3.jpg)

©︎
©︎
©︎
©︎
3. Filtering with Semantic Segmentation
Road + Sidewalk
≥ 10%
Selected
Not selected
jopparn, modified (CC BY-SA)
jopparn (CC BY-SA)
lvl5, modified (CC BY-SA)
okadatsuneo (CC BY-SA)
Helps select pedestrian / driver viewpoint


# Page. 11

![Page Image](https://bcdn.docswell.com/page/G78D3Q8L7D.jpg)

©︎
©︎
©︎
Filtering Results
Metadata
Non-AI Image
Analysis
AI-based Image
Analysis
Selected
…
okadatsuneo (CC BY-SA)
@drivephotograph (CC BY-SA)
lvl5 (CC BY-SA)
jopparn (CC BY-SA)
Build a filtering pipeline for your purpose


# Page. 12

![Page Image](https://bcdn.docswell.com/page/L7LMVXRQJR.jpg)

Finding the Street-Level Images You Need
✓The diversity of images is a key feature of Mapillary
✓Every image is valuable to someone
Filtering helps you find images that fit
your purpose!