Research map / Vision-language and multimodal / Cross-modal learning
TriSim: Tri-Dimensional Similarity Modeling with Extreme Value Theory for False-Negative Mitigation in Remote Sensing Image-Text Retrieval
Chengyu Zheng, Hanzhang Lu, Jie Nie, Shan Du
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 ETSM (EVT-guided 3D similarity FNS identification) to the baseline (ablation) | TriSim · RSITMD | mR: 47.96 -> 49.86 | Improved |
| add IGDO (intra-modal saliency gain matrix) (ablation) | TriSim · RSITMD | mR: 49.86 -> 51.35 | Improved |
| use 3D similarity thresholding per quadrant instead of single cross-modal threshold (ablation) | FNS detection · RSITMD | mR: 48.54 -> 49.16 | Improved |
| use Mahalanobis plus EVT filtering instead of per-quadrant thresholds (ablation) | FNS detection · RSITMD | mR: 49.16 -> 51.35 | Improved |
| replace single cross-modal threshold with Mahalanobis plus EVT filtering (ablation) | FNS detection · RSITMD | mR: 48.54 -> 51.35 | Improved |
| add IGDO on text retrieval (ablation) | TriSim · RSITMD | T->I R@1: 24.61 -> 26.95 | Improved |
| add IGDO on image retrieval R@1 (ablation) | TriSim · RSITMD | I->T R@1: 27.66 -> 26.55 | Hurt |
| TriSim vs best prior method on RSICD | TriSim · RSICD | mR: 36.99 -> 37.55 | Improved |
| TriSim vs best prior method on RSITMD | TriSim · RSITMD | mR: 50.22 -> 51.35 | Improved |
| vary quantile qu before last layer (0.8, 0.9, 0.95) (ablation) | ETSM · RSITMD | mR: peak 51.35 at 0.9 | Mixed |
| vary beta before last layer (0.05, 0.1, 0.2) (ablation) | ETSM · RSITMD | mR: peak 51.35 at 0.1 | Mixed |
| vary pg before last layer (0.05, 0.1 and higher) (ablation) | ETSM · RSITMD | mR: peak 51.35 at 0.1 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MaPLE — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 1 shared result
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
- Robust Remote Sensing Image-Text Retrieval with Noisy Correspondence (CVPR 2026 · similarity 0.88)
- MMM-RS: A Multi-modal, Multi-GSD, Multi-scene Remote Sensing Dataset and Benchmark for Text-to-Image Generation (NeurIPS 2024 · similarity 0.85)
- Cross-Modal and Uni-Modal Soft-Label Alignment for Image-Text Retrieval (AAAI 2024 · similarity 0.83)
- GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution (NeurIPS 2025 · similarity 0.83)
- SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing (AAAI 2024 · similarity 0.83)
- Accept the Modality Gap: An Exploration in the Hyperbolic Space (CVPR 2024 · similarity 0.83)
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.
