🌳 The taller the tree, the harder the fall. Determining tree height from space..
summary
Claims a deep‑learning stereo pipeline implemented in PyTorch produces canopy height maps from sub‑meter stereo satellite pairs that substantially outperform semi‑global matching and approach LiDAR‑derived canopy height models when trained on co‑registered LiDAR labels. It explains creating labels by projecting LiDAR into image pairs, handling terabytes of 40k×40k tiles at scale with Ray and Prefect, and guarding against hallucinated depths; the production product is the vegetation‑risk service Treeline.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:25Tree height importance
- 3:29Traditional measurement methods
- 6:41Stereo satellite imagery
- 10:11Challenges with stereo vision
- 13:01Deep learning benefits
- 16:30Training data creation
- 20:01Model architecture
- 22:29Scaling inference
- 23:54Key conclusions
- 24:35Q&A
related talks
36mWatching Millions of Trees with Python: For Impact and Profit
28mGeoAI: Evolution of AI-Based Tree/Vegetation Detection Models Using Imagery
48mTaking the planet’s pulse: AI to empower adaptation to global change
41mAbby Seneor | AI Track | Compass AI & Data Summit 2025
30mHow to Empower Utility Vegetation Management
36mDeep Learning - Segmentation, Identification and Classification
42mUsing Generative AI and Foundation Models to Predict Above Ground Biomass
32mMeasuring Vegetation Health to Predict Natural Hazards