LoCA: two-stage pipeline adds multi-modality to geospatial location encoders
Panagiotis Koromilas · hf · 2026-10-26
LoCA, shared on Hugging Face, is a two-stage pipeline that enables multi-modality in geospatial location encoders.
- Stage 1: Train co-aligned multimodal image encoders (Sentinel-2 multispectral + Sentinel-1 SAR) with a DHEL loss to build a shared embedding space.
- Stage 2: Freeze the image encoders as high-rank targets, attach them via adapter MLPs, and train location and metadata encoders with a multi-view contrastive loss (MV-DHEL).
- Inference: Coordinates feed a frozen location encoder plus linear probes to predict land cover, wildfire, climate, socio-economic and biodiversity targets, outperforming existing methods across diverse geospatial tasks.
More from Research
- ETH Zurich paper: dynamic evaluation allocation matches 10x-budget model rankings — zouharvi · 2026-10-26
- DuoLM: Fast LLM-Based Personalized Recommendation via Compact Encoder-Ranker — GoogleDeepMind · 2026-10-26
- DISC Learns to Combine Time-Series Dataset Similarity Metrics for Transfer Prediction — Yen-Ku Liu · 2026-10-26
- Adobe's SliderEdit++ Improves Slider-Based Image Editing by 18.5% with New SlideBench — Arman Zarei · 2026-10-26
- DeepMind Builds Auditable AI Text Detectors with Sparse Autoencoders — GoogleDeepMind · 2026-10-26
- CoRE Framework Learns Rule Routing for Knowledge Retrieval Without Supervision — Zhisheng Qi · 2026-10-26