Research map / Detection and segmentation / Open-set and few-shot detection
Generalized Decoupled Learning for Enhancing Open-Vocabulary Dense Perception
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 |
|---|---|---|---|
| integrate VFM distillation with region V-L alignment in one branch | region V-L alignment baseline · COCO region classification (Thing) | mAcc: 69.5 -> 65.6 | Mixed |
| decouple content and context objectives into separate branches | integrated learning · COCO Thing and Context59 | mAcc / mIoU: 65.6 -> 75.0 mAcc; 32.4 -> 35.3 mIoU | Improved |
| add content distillation (ablation) | F-ViT dense features · OV-COCO | mAP: 17.5 -> 37.6 | Improved |
| add content distillation (ablation) | F-ViT dense features · Training-free OVSS VOC21 | mIoU: 43.4 -> 42.3 | No clear effect |
| add context distillation (ablation) | F-ViT dense features · VOC21 and COCO-Obj | mIoU: 43.4 -> 59.7; 20.6 -> 36.4 | Improved |
| add context distillation to content distillation (ablation) | DeCLIP · OV-COCO | mAP: 37.6 -> 41.1 | Improved |
| add region correlation constraint (RCC) (ablation) | Content+Context DeCLIP · average score | average: 45.8 -> 46.6 | Improved |
| add SD-guided semantic completion (SD-GSC) (ablation) | Content+Context DeCLIP · average score | average: 45.8 -> 48.0 | Improved |
| add both RCC and SD-GSC (ablation) | Content+Context DeCLIP · average score | average: 45.8 -> 48.7 | Improved |
| use SD-GSC instead of direct feature concatenation of VFM and SD (ablation) | DeCLIP · VOC20 | mIoU: 75.2 -> 85.3 | Improved |
| use SD-GSC instead of direct alignment of SD self-attention (ablation) | DeCLIP · Context59 | mIoU: 32.0 -> 41.6 | Improved |
| fine-tune more CLIP attention blocks (ablation) | DeCLIP · OV detection and segmentation | performance | 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.
- ProxyCLIP — ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentation (ECCV 2024) · better on 1 of 1 shared result
- OpenMask3D — OpenMask3D: Open-Vocabulary 3D Instance Segmentation (NeurIPS 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
- DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception (CVPR 2025 · similarity 0.97)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.89)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.89)
- VITRIX-CLIPIN: Enhancing Fine-Grained Visual Understanding in CLIP via Instruction-Editing Data and Long Captions (NeurIPS 2025 · similarity 0.89)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.89)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.88)
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.
