Research map / Label-efficient and robust learning / Recognition and long-tail
TextFM: Robust Semi-dense Feature Matching with Language Guidance
Zhihao Zheng, Jinglun Feng, Nirav Savaliya, Zheng-Hang Yeh, Bo Lang, Mooi Choo Chuah
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 all components (LoRA VFM, priors, text-aware refinement, contextual alignment) (ablation) | FPN baseline · ScanNet (zero-shot) | AUC@5: 18.4 -> 22.7 | Improved |
| add LoRA-finetuned VFM features (ablation) | FPN baseline · MegaDepth | AUC@5: 55.5 -> 56.8 | Improved |
| add language-guided modules (TAR+CA) (ablation) | FPN baseline · ScanNet (zero-shot) | AUC@5: 18.4 -> 21.6 | Improved |
| add illumination-invariant physical priors (ablation) | FPN baseline · MegaDepth-N2N (zero-shot) | AUC@5: 42.6 -> 43.6 | Improved |
| use LoRA instead of frozen DINOv2 (ablation) | TextFM VFM · MegaDepth | AUC@5: 56.4 -> 56.8 | Improved |
| fully fine-tune DINOv2 instead of freezing (ablation) | TextFM VFM · MegaDepth | AUC@5: 56.4 -> 56.1 | Hurt |
| use CLIP ViT-L instead of DINOv2-L backbone (ablation) | TextFM · MegaDepth | AUC@5: 58.0 -> 57.1 | Hurt |
| use EVA02-L instead of DINOv2-L backbone (ablation) | TextFM · ScanNet (zero-shot) | AUC@5: 22.7 -> 22.5 | Hurt |
| use ADE20K 150-class vocabulary instead of learnable tokens (ablation) | TextFM · ScanNet (zero-shot) | AUC@5: 20.9 -> 22.7 | Improved |
| use ADE20K vocabulary instead of NYUv2 40 classes (ablation) | TextFM · ScanNet (zero-shot) | AUC@5: 21.9 -> 22.7 | Improved |
| use learnable prompt instead of fixed template (ablation) | TextFM · MegaDepth | AUC@5: 57.2 -> 58.0 | Improved |
| train at 800x800 instead of 480x480 (ablation) | TextFM · MegaDepth | AUC@5: 57.3 -> 58.0 | Improved |
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.
- TopicFM — TopicFM: Robust and Interpretable Topic-Assisted Feature Matching (AAAI 2023) · 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
- It's a (Blind) Match! Towards Vision-Language Correspondence without Parallel Data (CVPR 2025 · similarity 0.88)
- G^3-LQ: Marrying Hyperbolic Alignment with Explicit Semantic-Geometric Modeling for 3D Visual Grounding (CVPR 2024 · similarity 0.87)
- From Local Details to Global Context: Advancing Vision-Language Models with Attention-Based Selection (ICML 2025 · similarity 0.86)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · similarity 0.86)
- TextRefiner: Internal Visual Feature as Efficient Refiner for Vision-Language Models Prompt Tuning (AAAI 2025 · similarity 0.86)
- Mamba as a Bridge: Where Vision Foundation Models Meet Vision Language Models for Domain-Generalized Semantic Segmentation (CVPR 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.
