311
Deep learning CV and Earth Observation data. What satellites can see? Monitoring of Environment, Land management; Urban planning and Mapping, Mapflow.ai etc. Сontact us at hello@geoalert.io or @godnik0
Recently the most significant open buildings dataset was released by Technical University of Munich— “#GlobalBuildingAtlas”. It outperforms all other datasets (Google Open buildings, 3D-#GlobBFP) in the number of buildings (2.75 billion) and, in our estimation, also outperforms them in data quality.
As we at GeoAlert have been working on the benchmarks for our #Mapflow buildings model we used the opportunity to test the open data and contribute our validation to the work of the community. We focused this testing on the China urban area as there is a lack of validation data available for China.
In addition, we have published all the benchmarks, notebooks, and the QGIS project in the repository for easy visualization and reproduction.
It's been a while since we posted any research works on our blog. Read more. 😀
28 марта пройдёт русскоязычная открытая онлайн-конференция про ГИС и геотехнологии — Мапакон 2026! Мы планируем подготовить мастер-класс по AI-картографии (и да, мы специально не используем базз-термин GeoAI, потому что такого приложения нет =) Расскажем почему на Мапаконе)
Читать полностью…
#Mapflow usage tips. #Building Regularization option considers the rounded shapes (based on the corresponding structures, which got a well approximated mask)
Читать полностью…
Wrapping up the year with one more update — the new version of Mapflow QGIS plugin is out. 🎉
3.5.0 (experimental)
- Add translations to Chinese, Spanish and Portuguese
- Save results of all geometry types as one file's layers (GPKG)
- Tree crowns' style for 🌲 Forest and trees
- Water objects' style for land use (based on the [multi] Buildings + Roads + Forest)
Don't forget to update! (all versions)
#Mapflow AI-mapping tips 👉 If you use the 🌲 Forest model with "tree crowns" option, we recommend inputting imagery data of 0.3m/px. Which is equal to ~ zoom 19 of the imagery basemaps.
"Tree crowns" is the separate model and it would perform much better on zoom 19, while the basic model of the Forest mask is not so sensitive to the range of resolution.
For more detailed information on the model requirements, check out our reference documentation, or ask the AI-chatbot available in the Mapflow (yes, we know that none of us likes to read the documentation).
From the user #feedback. Dense residential area in Iraq.
Читать полностью…
We’ve spent a good amount of time striving to find the best possible simplification for Buildings model that would produce the map-friendly polygons and align with user expectations. And we kept debating whether it’s better to call it “simplification” or “polygonization”. )
🚀 As the Mapflow was designed with the idea of striving for UI simplicity, we came up with a compromise to settle on two options instead of one.
Read more in our blog 👀
👉 We’ve also adjusted the pricing structure: the model price has increased, while simplification now costs slightly less.
👆 Use NAIP imagery as a custom data source in Mapflow.
What’s good:
NAIP (National Agriculture Imagery Program) offers nationwide aerial imagery with an effectively public-domain license, so you can freely use it, integrate it into your projects, or combine it with other data.
What’s not so good:
The imagery is downsampled to ~0.6 m and is not very recent—updates typically occur every 2–3 years. The latest available dataset is from 2022, as confirmed via ESRI’s metadata service.
How to connect:
Paste the NAIP TMS link into Custom URL in Mapflow Web or add it as a new provider in the Mapflow for QGIS plugin settings. After that, you’re ready to run your processing.
See more examples of custom data sources in our user guides.
Mapflow Imagery Search – how it works for the monitoring projects.
1) Add AOI
2) Get notified about the updated search results
3) Run the analytical workflow using the new satellite image
🚀 We’re hiring! Join the Geoalert Team as a CV/ML Engineer
Build and deploy real-world computer vision solutions — while crossing research gaps, walking the uphill of uncertainty, and wrestling with ever-mysterious metrics. A challenging path, but full of growth
You’ll:
• Train & optimize DL models
• Design data pipelines
• Develop pre/post-processing
• Improve model performance
We expect:
• 2–3 yrs CV experience
• PyTorch, OpenCV, Scikit-image
• Detection/segmentation/depth tasks
• ML fundamentals, Linux, Git, Docker
🌍 Remote | Impactful work | 💡 Tech-driven team
Contact us and send your CV*CV at hello@geoalert.io
A little late for #30DayMapChallenge - My Data, but sharing our #Mapflow usage statistic for last month.
The trends of the usage by countries and the low level data - which it the number of imagery analysis processings made by Mapflow users displayed on the global map.
For the data we use Postgres/Postgis and for the statistics and charts we use Apache Superset.
From the Mapflow user's #feedback. Multi-model processing. No classification options. Territory - Iraq.
Читать полностью…
Just a random photo from a Tel Aviv gathering. (Occasional workspace 😉)
Читать полностью…
Mapflow-QGIS plugin 3.3.0 is out. 🚀
What’s new:
1) New buttons in the Processing menu: restart (the failed one), duplicate processing params
2) My Imagery: rename images
3) Go to the corresponding image in "My Imagery" from the Processing details (before it was quite tricky to find out the image if you have a lot of them)
4) New styles for the "Open data download"
Don't forget to update when starting QGIS. ✍️
For those of our followers who might not be familiar with this project – we started developing the QGIS plugin with the primary purpose of showcasing the Mapflow API and getting a smoother integration of it for those who are used to working with GIS software. Currently, the plugin has over 160,000 downloads from the official repository, and users have created videos and other content teaching how to use it for mapping and analytical projects.
Join the community and follow our progress on GitHub — post an issue, suggest a feature, or contribute to the code:
👉 github.com/Geoalert/mapflow-qgis
Let's meet at INTERGEO #Expo and Conference 2025! It's very inspiring to attend an event of this scale.
#Geoalert CEO Georgy Potapov will be at the Expo in Frankfurt. We look forward to meeting with our partners and to get introduced to the new ones.
Note that GeoAlert software is featured at #geocloud.work booth, who are our fellow partners.
Мапакон 2026 — про геотехнологии в реальной жизни
Гео — это не только ПО и инструменты. Это бизнес-решения, команды, продукты и обучение специалистов. Именно поэтому мы ценим опыт на стыке:
— геоданных и бизнеса,
— технологий и продуктовых решений,
— практики и образования.
📍Если у вас есть живые кейсы, которыми вы готовы поделиться с профессиональным сообществом — сейчас самое время подать заявку.
А если вы работаете в гео-компании, вы можете предложить Мапакон как площадку для экспертного присутствия: через доклады, партнёрские форматы и диалог с аудиторией.
Приём заявок на доклады — до середины января.
Подробности для компаний — на странице партнёров.
🔸На сегодня уже подтвердили спонсорское участие компании GeoAlert и Геоинтеллект
#mapacon
The funny part for us in summarizing a year is to display our metrics on the map )
Recently, we defined our Noth Star metric as the number of credits spent by the users of our platform.
Why should we use credits instead of area (e.g., the common pricing unit for satellite imagery providers and resellers) or number of processings made by users?
Credits help us to adjust the pricing for data and services and represent the normalized value of processing results.
Different models and source imagery may have different cost in credits even for the same area.
⭐ North Star metric = Total Credits spent by Users
This metric reflects:
- How much work the product has delivered
- How much value users have received
- How deeply users engage with different models
- How product adoption grows over time
- How the overall product revenue increases
This metric always reflects the value that users receive, not the money the business earns.
Why can’t we use only paid processings? 💰
Free processings represent activation - the first step toward monetisation;
They reflect the core value: the user starts processing and receives a result;
When free processings grow, paid conversions typically grow as well.
Many thanks to a good old fried of us from "Lab IV" (small consultancy firm based in Berlin), who helped us to get a motivation for the metrics discussion and development, and facilitated the internal team communication.
Thank you for being a member of the #Mapflow users community! We're excited to keep growing the product and to learn from your feedback and from the use cases of our customers. Happy mapping in the upcoming year!
Oh... we love maps. Hense we would like to share some meaningful Mapflow metrics that we started collecting this year.
Isn't it beautiful? At least the data-mapping approach, if not the metrics themselves. 🤩
A New Year turnover in your career 👆 – consider applying to the new open role in our team. We have a non boring legacy of the AI models, and a secret Santa with the challenging new projects. 🎅👨💻
Читать полностью…
🚀 🚀 We’ve just updated one of our most popular AI mapping models — "🌲 Forest and Trees"
The default workflow now includes two output options for tree crowns: polygons and points.
The new points option extracts individual tree crowns as point features enriched with area and radius attributes. This significantly reduces output file size (by 2× or more) while meeting common mapping requirements where trees are represented as simple geometry primitives.
Despite the simplified geometry, the data is still based on real tree crown estimates, allowing you to apply meaningful, data-driven styling in your maps.
Try the new tree extraction options in your own mapping designs.
NEW Satellite Imagery Search in #Mapflow. How it works in two lines:
1) You get the notifications about the imagery search updates (which runs in the background and constantly updates you about new data)
2) You check it and see the list of the new images found by the intersection with your area of search
Try it yourself. Whenever you used to struggle with the satellite imagery catalogues, this service aims to disrupt your user experience.
🔈 Join our next webinar, where we will introduce the new powerful tool for imagery search and updates. Learn how to select different data sources and run your analytical workflow.
Demo and Q&A! 😎
Default #buildingfootprints model in #Mapflow - improving recall and simplification.
Sample images (Before – After): US, Australia, Africa.
Try it for free at mapflow.ai
Why #Mapflow Search is not yet another commercial imagery marketplace with carts and checkouts? 🙈
Read the full story on our blog or just watch this 1-min video and try it yourself.
(In Russian) Мы давно планировали и недавно завершили интеграцию данных Роскосмоса. Эти снимки наиболее релевантны для территории России. Два спутника, данные которых доступны для анализа с помощью моделей Mapflow, обеспечивают высокую производительность ДЗЗ. Проверить наличие снимков можно через поиск на портале провайдера или прямо в сервисе #Mapflow.
Как работает поиск:
1. Загрузите или нарисуйте область интереса.
2. Выберите данные — можно выбрать часть области, а оставшуюся сохранить в режиме поиска и получать уведомления о появлении новых снимков прямо на платформе.
Как запустить обработку:
3. Выберите подходящий снимок.
4. Перейдите к выбору модели и параметров, затем запустите анализ.
Качество этих данных уступает коммерческим источникам вроде Mapbox или ArcGIS, однако они позволяют работать с более свежими или историческими снимками. Если вам нужен доступ к этим данным — свяжитесь с нами, чтобы подключить провайдера (доступен в тестовом режиме).
Join our #webinar tomorrow: using #Mapflow CLI tools and My Imagery with #QGIS! Register here 👈
Читать полностью…
🚀 We’ve published the Mapflow CLI tool!
It’s designed for those who interact with the #Mapflow API using #Python — perfect if you’re automating data pipelines or processing large sets of imagery.
Actually we were asked by users how to run automated data workflows with the API — and we were surprised to learn that some companies still do this manually! So here’s what the CLI can do for you:
👉 Data handling: No need to upload images one by one — just point the script to your local folder.
👉 Processing automation: Select a collection, run a model over its full extent or a specific polygon within
Note that as you have your images stored in the Mapflow My Imagery you can rename them, link to another collection - these operations can be done independently to organise your data.
We wrapped up some other methods considered to be useful, such as getting the list of available models with:
python -m scripts.processing models
We redesigned the sidebar—now there is a single dropbox for uploading your AOI polygon or GeoTIFF file. The latest calculates the image extent; you can choose it or apply the vector polygon instead. ✍️#Mapflow #tips
Читать полностью…
It was a pleasure to meet our fellow partners at Intergeo Expo. As a brief overview, I'd rather say that this year there are even more hardware-based verticals with even more stunning equipment that inhouse every possible geo-technology in them, from lidars to aerial survey and AI analytics - this is on one end, and outsourcing & outstaffing companies doing whatever you want in the field of Geospatial - on the other. There are very few product-focused companies specializing in software design or analytics, so it's hard to determine if they have any clients in this area. (The most successfully positioned and focused company is supposed to be the one designing and producing the lasting cases for geodesic equipment.)
It seems that German vendors use the Expo to meet their key government customers and offer them a cup of coffee. The last thing is some technical or data opportunities they would care about.
Nevertheless, it was my good fortune that I was accompanied by a friend Ilya Vladimirskiy, who knows the broad data market, based in Berlin, and wants to help us build some connections.
Besides browsing the Expo, we met with our partners - the Chinese satellite data providers, the cloud marketplace for geospatial software geocloud.work, and the users of our software.
BTW, I've come across Plan4Better, who started their journey by spinning off a university open-source project, around the same time we started #GeoAlert.
From the Mapflow user's #feedback (positive) – Building footprints with classification by typology.
Now it's much easier to read the classification in the preview with the map legend.