Research map / Detection and segmentation / Open-set and few-shot detection
TerraMesh-Masks: Open‑Vocabulary Segmentation for Earth Observation
Benedikt Blumenstiel, Hugues Devimeux, Johannes Jakubik, Konrad Schindler
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces an open-vocabulary segmentation effort for Earth observation, motivated by the lack of large-scale EO-specific segmentation datasets; details beyond the truncated abstract are unavailable.
Problem
Open-vocabulary segmentation transfers poorly to Earth observation data because of differing semantics, scales, sensing modalities, and geographic context, and large-scale EO-specific segmentation datasets are lacking.
Approach
Builds an EO-focused open-vocabulary segmentation setting and dataset (TerraMesh-Masks) to address the absence of large-scale EO segmentation data; the specific method is not stated in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train on TerraMesh-Masks (TerraMind-OVS-L) instead of using general-purpose OVS models | TerraMind-OVS-L · TerraMesh-Masks-Eval | IoU_pos: 16.29 (DINOv3.txt) / 11.33 (SAM-3) -> 35.18 | Improved |
| Scale TerraMind-OVS from small to large (ablation) | TerraMind-OVS · TerraMesh-Masks-Eval | IoU_pos: 26.52 -> 34.87 | Improved |
| Use S2-only trained models instead of multimodal-trained (ablation) | TerraMind-OVS-L · TerraMesh-Masks-Eval | IoU_pos: 34.87 -> 34.96 (RGB) | No clear effect |
| Add multimodal (S1, DEM) input at inference (ablation) | TerraMind-OVS-B · TerraMesh-Masks-Eval | mIoU | No clear effect |
| Use S2 multispectral instead of RGB input (ablation) | TerraMind-OVS · TerraMesh-Masks-Eval | mIoU (Forest & Wetland, Barren terrain) | Mixed |
| Apply NDVI agreement filter at pretraining (keep score at least 0.4) (ablation) | TerraMind-OVS · TerraMesh-Masks pretraining | training signal | No clear effect |
| Use MS-CLIP prompt templates instead of ImageNet templates (ablation) | DINOv3.txt · TerraMesh-Masks-Eval | mIoU | Hurt |
| Train TerraMind-OVS-L on TerraMesh-Masks, evaluated on SAR out-of-distribution | TerraMind-OVS-L · CaFFe (OOD SAR) | mIoU: 15.32 (DINOv3.txt) vs 8.24 | Hurt |
| Train TerraMind-OVS on TerraMesh-Masks, multi-label evaluation | TerraMind-OVS-B · reBEN-7k test | mIoU: 3.52 (DINOv3.txt) -> 13.75 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SAM-3 — SAM 3: Segment Anything with Concepts (ICLR 2026) · better on 3 of 3 shared results
- SegEarth-OV — SegEarth-OV: Towards Training-Free Open-Vocabulary Segmentation for Remote Sensing Images (CVPR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Locate Anything on Earth: Advancing Open-Vocabulary Object Detection for Remote Sensing Community (AAAI 2025 · similarity 0.88)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.88)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.87)
- Terra: A Multimodal Spatio-Temporal Dataset Spanning the Earth (NeurIPS 2024 · similarity 0.87)
- SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · similarity 0.86)
- TEOChat: A Large Vision-Language Assistant for Temporal Earth Observation Data (ICLR 2025 · similarity 0.86)
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.
