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
Geoalert #Team at ICT week, Tashkent, past week.
Читать полностью…
The "Blood moon" over Petrovaradin fortress, Serbia. 2025-09-07 (photo by Geoalert's CTO Alex Trekin)
The moon appears red during lunar eclipses because the only sunlight reaching it is reflected and scattered through the Earth’s atmosphere. 🌛
We are starting our webinar in 15 minutes. Join us for the live demo and teasing about new options related to #Mapflow Buildings model and open data.
Читать полностью…
From the #Mapflow user's feedback. Combined models applications to the satellite images of the rural area in China. 🤩
Читать полностью…
First #feedback about Mapflow #Buildings + Heights model, using Global Mosaic 2022 data source. Thanks for sharing, good so far. 😃
Nevertheless, the input image quality always impacts the model's performance. And there is room for improvement of the model's stability. Stay tuned for the model's update and try it out.
🚀 Mapflow Buildings Model: Now with Estimated 🔝Heights (Beta)
We’ve rolled out a major update to the default Mapflow Buildings model: you can now estimate building heights directly from imagery. This new feature leverages a dedicated regression-based model that infers height using visual indicators such as shadow length and visible wall segments.
The result is what they call 3D building footprints—where the building's contour is projected to ground level instead of the roof outline. This is especially useful for oblique imagery, where roofs often appear shifted.
You can enable this feature in the default workflow. It’s available across all interfaces: Mapflow Web, QGIS plugin, and Mapflow API.
TL;DR: What to Know About 3D Building Footprints
– Lidar and stereo-pair imagery offer higher precision for 3D modeling, but they require specialized, often costly data and processing pipelines.
– Mapflow’s new method trades a bit of precision for high scalability and minimal input requirements—ideal for large-scale applications where budget and speed matter.
– For enterprise use cases like telco infrastructure planning, we offer additional accuracy tuning using satellite metadata (e.g. sensor angle). It's like restoring the building's geometry out from the known projections and helps us to achieve a mean absolute error (MAE) just over 3 meters—about one floor of a typical residential building.
This approach works without stereo input or metadata, making it especially cost-effective and accessible.
While we continue gathering benchmark data and performance metrics, we will appreciate your feedback. Test it and find out the application to your particular tasks. If the output doesn’t meet expectations, let us know—we monitor ⭐ Mapflow rates and comments closely and offer refunds for poor results. Your input helps us prioritize improvements and tailor the tool to your needs.
Complex shape simplification - DYNAMIC_GRID, It was released a few months ago within the #Building footprints workflow in #Mapflow and replaced all the shape-based simplifications beyond the CIRCLE and RECTANGLE.
#Research #Mapping
#Constrcutions mapping and change detection
Combining multiple images with Mapflow construction detection enhances change detection for monitoring construction and urban development over time. A potential workflow involves:
- selecting a starting point (e.g., GM 2022, 2023, or 2024) and detecting constructions using Mapflow;
- ordering newer imagery for detailed analysis if the existing mosaic is insufficient.
This allows application of models like Buildings or Vegetation to analyze the context of changes.
Note. We are developing a fully integrated workflow for online image discovery and ordering.👆
We are looking for CV / ML engineer 🔥
Help build and deploy computer vision solutions for real-world challenges.
Train and optimize deep learning models for production
Design data pipelines and manage annotations
Develop pre/post-processing algorithms
Monitor and improve model performance
Write production Python code and collaborate with developers
2–3 years of CV experience
Strong skills in PyTorch, OpenCV, Scikit-image
Experience with detection, segmentation, or depth tasks
Solid ML foundations and problem-solving mindset
Proficiency with Linux, Git, Docker
Real product impact
Fast, flexible, and tech-driven team
Fully remote work with cutting-edge tools
We've updated the #Mapflow Processing API documentation. You can find all processing params in the reference section.
The API methods for running processing with the "data source" have been updated to allow for more specific use based on the type of data source you utilize: the default data provider, a custom data provider specified by a URL, or local images organized in the #imagery collections through Mapflow My Imagery. The older version of the API will still be supported in the meantime.
One application of the new combo scenario is mapping tiny urban areas where there is little to no data in OpenStreetMap and other open sources. As the new settlements pop up, they should be presented visually and included in the topographic plan. Note that you can get the relevant results in a few clicks with the help of Mapflow AI.
Читать полностью…
New webinar video is available on our channel. Like it if you find this presentation useful. 👍
Читать полностью…
We set the updated date-time for our next #webinar introducing to #Mapflow-#QGIS 3.2.0 and to new default scenario for landuse mapping. Save your spot!
Читать полностью…
The new Mapflow for QGIS version 3.2.0 is out. The major update is adding the Projects view. Organising the projects was not so easy for plugin users as they needed to switch between the Settings and Processing tab. Now the Projects are added as nested up-level table - so the plugin's UI looks more similar to the web. User can go up to the Projects table, sort columns and filter projects by name. This feature was primarily requested by those who create multiple projects and use the plugin as the main tool for coordination in the Mapflow teams.
However, it is useful for individual users who have many processings to organise them and group them in Projects. Free users are not limited to the number of Projects. So don't forget to upgrade the plugin at the next launch of QGIS and share your feedback with us. We appreciate it.
Other updates in 3.2.0:
- Preview imagery collections
- Create AOI from a single image or collection extent
- Confirm before starting processing
- Filter imagery search by data provider
https://youtu.be/srFCEiH-YZ0
BTW The music used in the video was written by one of our colleagues.
New Feature: Layer Switching & Legend in Mapflow Viewer
You can now switch layers directly in the Mapflow map viewer. Use the menu button (☰) in the top-right corner of the map to open the layer panel and legend.
Since model outputs are generated as a single GeoJSON, the legend helps you distinguish between object classes such as Buildings, Roads, and Vegetation when using the combo scenario. This makes it easier to preview classification results—whether classifying buildings by typology or vegetation by height.
(In the GeoJSON, classification values are stored in the class_id property.) ⭐️
When comparing #AI-powered mapping with open data, one of the most common questions we hear is: what’s the real value of imagery analysis and feature extraction compared to openly available datasets?
In practice, it depends on your needs:
Image alignment – if you already have your own imagery and want features to align with it (the value is in the tool, not just another dataset).
Data updates – if you need the most recent imagery to detect changes or trust results based on a specific source.
Still, open mapping data is a great reference point and complement to AI-based projects.
🚀 New release:
We’ve added a new default scenario to download data directly from #OpenStreetMap and #OvertureMaps – two most trusted open mapping data sources.
Workflow is simple:
1. Search, draw, or upload AOI
2. Select a background image (Mapbox Satellite is free, others may require credits)
3. Choose the open data object types (4 available now, more coming)
4. Run the workflow & preview results on the map
Now you can combine open mapping data with AI mapping results in a single project.
Regarding the Pricing:
We set minimum price = 1 credit / sq. km / feature type
Not for large-scale downloads, but perfect for combining open data with imagery basemaps + AI mapping.
🌍 Examples:
1. Airport in Australia – #Mapflow struggles with terminal structures, but OSM maps them well → just import from OSM.
2. Area in Texas – U.S. datasets are strong, but forestry areas are rough → use AI detection to refine and update.
💡 These are just a few examples. We’d love to hear your ideas, comments, and requests on using the new scenario!
As August is vacation time in many companies, we share some photos of our teammates on their adventures across different countries (Part 1). How many countries can you name? #GeoQuizz #workation #Geoalert #Team
Читать полностью…
See the demo of the Buildings model with "Height estimation" based on Tangshan (China) sample. Check the results in the demo project.
Читать полностью…
Occasional workspace #Workation #Kakheti #Georgia
Читать полностью…
In communication with #Mapflow users, we often come across those who upload the data they obtained or purchased from the external imagery supplier to analyse it with AI #Mapping.
If that’s your case, here are a few tips and links that might help you prepare and manage your data more efficiently:
✅ Optimize image size
Reducing image size helps minimize upload time and storage usage. Here's how to do it:
👉 https://docs.mapflow.ai/userguides/howto.html#how-to-optimize-large-image-files
✅ Use the QGIS plugin for mosaic uploads
If you're using QGIS, you can upload ortho imagery one by one (yes, ortho imagery is usually supplied split into parts) directly into your Mapflow #Imagery collection. This allows you to process them as a single mosaic.
👉 https://docs.mapflow.ai/userguides/my_imagery.html#my-imagery-in-qgis
✅ Want to pack multiple models into a single workflow?
Contact support (or use the chatbot) if you'd like to combine several AI models into one mapping workflow. This enables you to run multiple models (e.g., 🏠 Buildings + 🌲 Forest) together and apply #topology correction to the results, avoiding overlaps between different feature types.
If you have questions—or tips of your own about handling ortho imagery—we’d love to hear from you! (Example below: a user-rated map.)
When, because of the layers order in QGIS, it looks like a feature map with zero-shot segmentation... (yet the features are the results of segmentation powered by Mapflow)
Читать полностью…
🌍 Global mosaic updates 🔥
Starting from June we provide Global mosaic updates by request. Current updates are available to the year 2024. And the more recent imagery is expected to arrive by the end of the year.
The old version of the Global mosaic of high-res imagery (0.75–0.5 m/px) you can try in Mapflow as of the year 2022. Preview is limited to zoom 15.
Note that you can also upgrade your account to access World Imagery, a global coverage composed of high- and medium-resolution #satellite imagery and aerial imagery, hosted by Esri.
From the user' #feedback. Roads in the mountains area, Colombia
Читать полностью…
New #QGIS styles for the #Mapflow Forest and Trees global model. Don't forget to update to v.3.2.0 👆
Читать полностью…
From the user's #feedback. Using a combo scenario with selected options. (Thanks to those who share their evaluation and leave comments!)
Of course there are inaccuracies in the mapping results that can be fixed manually in a small fraction of time needed to draw all these things from scratch.
Give it a try and share your rating.
Geoalert #team is growing - we have been joined by Khristina, who'd step into a role as GIS analyst. 🤩👩💼👨💻
Читать полностью…
New Default #Mapflow Scenario: Combined Mapping of Buildings, Roads, and Vegetation.
We've introduced a new default scenario in Mapflow that combines three models in a single workflow — designed to map buildings, roads, and vegetation all at once. The result is a topology-corrected vector layer that's ready to use.
Simply select the new scenario from the menu and streamline your mapping process.
Give it a try, rate the results and share your feedback. We’re always listening.
Users' feedback from Madagascar. 🇲🇬 Thank you to those who share.
Читать полностью…