Research map / Probabilistic and scientific modeling / Variational and generative sampling
GeoCFM: Positive-Only Conditional Flow Matching for Mineral Occurrence Sampling
Moshe Eliasof, Eldad Haber
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace UNet-Seg score-map with GeoCFM conditional flow-matching sampler | UNet-Seg · Synthetic magnetics-geochemistry benchmark | Chamfer distance (px): 29.46 -> 9.37 | Improved |
| replace UNet-Seg score-map with GeoCFM conditional flow-matching sampler | UNet-Seg · USGS Earth MRI (tile-disjoint test split) | Chamfer distance (px): 46.81 -> 12.30 | Improved |
| condition on geo-images (GeoCFM) instead of unconditional global KDE | Global KDE · Synthetic magnetics-geochemistry benchmark | Chamfer distance (px): 41.16 -> 9.37 | Improved |
| condition on geo-images (GeoCFM) instead of unconditional global KDE | Global KDE · USGS Earth MRI (tile-disjoint test split) | Chamfer distance (px): 66.35 -> 12.30 | Improved |
| use positive-only flow matching instead of pseudo-negative GBDT score map | GBDT · USGS Earth MRI (tile-disjoint test split) | top-5% hit rate: 0.102 -> 0.250 | Improved |
| use 5 Euler steps instead of 50 (ablation) | GeoCFM · Synthetic magnetics-geochemistry benchmark | Chamfer distance (px): 9.37 -> 9.30 | No clear effect |
| use 5 Euler steps instead of 50 (ablation) | GeoCFM · USGS Earth MRI (tile-disjoint test split) | Chamfer distance (px): 12.30 -> 12.00 | No clear effect |
| use 5 Euler steps instead of 50 (ablation) | GeoCFM · Synthetic magnetics-geochemistry benchmark | F@5: 0.634 -> 0.580 | Hurt |
| use 5 Euler steps instead of 50 (ablation) | GeoCFM · Synthetic magnetics-geochemistry benchmark | ms per patch: 38.47 -> 19.79 | Improved |
| train with 50% of positives instead of 100% (ablation) | GeoCFM · Synthetic magnetics-geochemistry benchmark | Chamfer distance (px): 9.38 -> 9.60 | No clear effect |
| train with 50% of positives instead of 100% (ablation) | GeoCFM · USGS Earth MRI (tile-disjoint test split) | Chamfer distance (px): 12.35 -> 12.46 | No clear effect |
| increase sampling budget N from 100 to 2000 (ablation) | GeoCFM · Synthetic magnetics-geochemistry benchmark | KDE NLL: 11.79 -> 10.56 | Improved |
Closest papers by meaning
- Towards Automated Petrography (NeurIPS 2025 · similarity 0.83)
- A Unified Framework for Forward and Inverse Problems in Subsurface Imaging using Latent Space Translations (ICLR 2025 · similarity 0.82)
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.82)
- CAR-Flow: Condition-Aware Reparameterization Aligns Source and Target for Better Flow Matching (NeurIPS 2025 · similarity 0.81)
- Coreset-Induced Conditional Velocity Flow Matching (NeurIPS 2026 · similarity 0.81)
- Fast samplers for Inverse Problems in Iterative Refinement models (NeurIPS 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
