From Maps to Models: An End-to-End Journey in Geospatial Machine Learning with Python
Talkat PythonAsia 2026 · on Saturday, 2026-03-21· at 13:15· Yuchengco Hall 4th Flr. Y409 (Workshop Room 2)·
summary
A practical end to end pipeline shows how raw satellite rasters and vector building footprints become training tiles, a U‑Net (ResNet encoder, ImageNet weights) trains on 512×512 tiles with 30–40% overlap, and inference produces building masks that still miss examples after a brief 10‑epoch run; notebooks run in Google Colab and require GPU. Tools named include leafmap, geopandas, rasterio and goiip; resolution rules are explicit (10 m raster → ~5 m usable accuracy, rounding coordinates to 3 decimals can shift ~100 m).
AI-written summary · about
chapters (11)
- 0:00Workshop setup
- 2:31Introduction to geospatial machine learning
- 8:13Raster vs vector data
- 14:11Building data pipeline
- 29:39Preprocessing geospatial data
- 36:55Training segmentation model
- 59:19Geospatial data accuracy challenges
- 1:08:27Beginner geospatial projects
- 1:13:00Q&A
- 1:21:10Philippines data sharing
- 1:28:38Top geospatial machine learning applications
related talks
17mGeoAI: a feature extraction tool set using Tensor Flow and Keras...
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
33mWhat and where? - Machine learning for geospatial image analysis
41mLarge Scale Field Delineation
26mGeospatial Machine Learning at Scale
36mClassification of Satellite Imagery
46mDeep Learning for Satellite Image Processing, Lessons Learned
31mUnpacking the geospatial engineering toolbox – an overview of data science techniques for spatial da