Research map / Large language models / AI and society
EoS-FM: Can an Ensemble of Specialist Models act as a Generalist Feature Extractor?
Pierre Adorni, Minh-Tan Pham, Stephane May, Sébastien Lefèvre
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 |
|---|---|---|---|
| add BatchNorm on encoder features (ablation) | EoS-FM · Feature normalization, 6-dataset average | val mIoU: 57.01 -> 65.15 | Improved |
| add LayerNorm on encoder features (ablation) | EoS-FM · Feature normalization, 6-dataset average | val mIoU: 57.01 -> 62.70 | Improved |
| add BatchNorm on encoder features (ablation) | EoS-FM · CropTypeMapping | val mIoU: 15.14 -> 60.70 | Improved |
| add BatchNorm on encoder features (ablation) | EoS-FM · MADOS | val mIoU: 66.06 -> 62.42 | Hurt |
| add BatchNorm on encoder features (ablation) | EoS-FM · HLS Burn Scars | val mIoU: 80.52 -> 84.41 | Improved |
| add BatchNorm on encoder features (ablation) | EoS-FM · AI4SmallFarms | val mIoU: 37.55 -> 39.56 | Improved |
| add BatchNorm on encoder features (ablation) | EoS-FM · SpaceNet 7 CD | val mIoU: 54.85 -> 54.76 | No clear effect |
| increase number of encoders from 1 to 4 (ablation) | EoS-FM · HLS Burn Scars | val mIoU: 66% -> 90% | Improved |
| increase number of encoders beyond 4 (ablation) | EoS-FM · HLS Burn Scars | val mIoU: stagnates around 90% | No clear effect |
| increase number of encoders from 1 to 17 (ablation) | EoS-FM · AI4SmallFarms | val mIoU: 35% -> 45% | Improved |
| prune ensemble to top-6 encoders (EoS-FM Small, 22M vs 76M) (ablation) | EoS-FM · Pangaea (full data) | Avg DTB: 4.58 -> 5.13 | Hurt |
| prune ensemble to top-6 encoders (EoS-FM Small, 22M vs 76M) (ablation) | EoS-FM · Pangaea (10% labels) | Avg DTB: 3.67 -> 7.78 | Hurt |
Closest papers by meaning
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · similarity 0.86)
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.86)
- No One Knows the State-of-the-Art in Geospatial Foundation Models (NeurIPS 2026 · similarity 0.86)
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.85)
- SkyMoE: A Vision-Language Foundation Model for Enhancing Geospatial Interpretation with Mixture of Experts (AAAI 2026 · similarity 0.85)
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.85)
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.
