GeoAI Tutorials - Building Instance Segmentation Step-by-Step with GeoAI

Unlock the power of instance segmentation for buildings using the open-source GeoAI Python package! 🏗️ In this hands-on tutorial, you'll learn how to train, run, and evaluate a deep learning model that can accurately detect and separate individual buildings, even when they're tightly clustered. From data preparation and training in Google Colab to inference, post-processing, and geometry analysis, this video walks you through everything you need to know — step by step. ✅ What you’ll learn: Semantic vs Instance segmentation explained How to train your own Mask R-CNN model Working with high-resolution satellite imagery Converting raster to vector footprints Filtering, visualizing, and regularizing building data Ideal for GeoAI developers, remote sensing professionals, and anyone in geospatial deep learning. Don't forget to subscribe for more GeoAI tutorials! Notebook example: https://opengeoai.org/examples/train_instance_segmentation_model 00:00 - Introduction to Building Instance Segmentation 00:44 - Why Instance Segmentation Over Semantic Segmentation 01:19 - Visualizing the Expected Results 01:47 - Accessing Sample Notebooks on OpenGeoAI 02:35 - Running on Google Colab with GPU 03:00 - Installing Required Python Packages 03:56 - Downloading and Preparing the Sample Dataset 04:46 - Visualizing Raster and Vector Data 06:00 - Differences Between Training and Test Data 06:37 - Creating Image Chips for Model Training 07:51 - Understanding Label Generation and Tile Output 08:54 - Training the Mask R-CNN Model 10:56 - Model Training Parameters Explained 11:41 - Monitoring Model Performance: IOU & Loss 12:58 - Saving and Locating the Trained Model 13:38 - Running Inference on New Dataset 14:42 - Adjusting Confidence Threshold for Accuracy 15:22 - Raster to Vector Conversion & Regularization 16:45 - Adding Geometry Properties (Area, Shape, Orientation) 18:10 - Filtering Small or Irregular Buildings 19:30 - Visual Comparison: Input Image vs Detected Buildings 20:50 - Post-processing: Color Maps & Clean-up 22:30 - Limitations & Real-world Considerations 23:20 - Final Thoughts & Performance Metrics 24:30 - When to Use Semantic vs Instance Segmentation 25:20 - Batch Inference & Multispectral Support 26:00 - Wrap-up & Tips for Better Results #InstanceSegmentation #GeoAI #DeepLearning #RemoteSensing #BuildingDetection #SatelliteImagery #PythonAI #MaskRCNN 📺 GeoAI Playlist: https://www.youtube.com/playlist?list=PLAxJ4-o7ZoPcvENqwaPa_QwbbkZ5sctZE 📘 Get my new Book - Introduction to GIS Programming: A Practical Python Guide to Open Source Geospatial Tools 👉 Amazon: https://amazon.com/dp/B0FFW34LL3 👉 Leanpub: https://leanpub.com/gispro 👋 Let’s Connect: YouTube: https://youtube.com/@giswqs LinkedIn: https://www.linkedin.com/in/giswqs Twitter: https://twitter.com/giswqs Facebook: https://www.facebook.com/groups/opengeos