Research map / Architectures and efficiency / Attention and transformers
UniRefiner: Teaching Pre-trained ViTs to Self-Dispose Dross via Contrastive Register
Congpei Qiu, Zhaoyu Hu, Wei Ke, Zhuotao Tian, Yanhao Wu, Tong Zhang
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 UniRefiner | EVA-CLIP 8B/14 · ADE20K linear probing | mIoU: 42.5 -> 51.9 | Improved |
| add UniRefiner | DINOv2 G/14 · ADE20K linear probing | mIoU: 49.1 -> 50.6 | Improved |
| add UniRefiner | SigLIPv2 So/16 · ADE20K linear probing | mIoU: 45.6 -> 49.8 | Improved |
| add UniRefiner | EVA-CLIP 8B/14 · NYUv2 depth linear probing | delta1: 77.9 -> 90.9 | Improved |
| add UniRefiner | EVA-CLIP 8B/14 · Open-vocab segmentation Cityscapes | mIoU: 11.4 -> 30.1 | Improved |
| add UniRefiner | SigLIPv2 So/16 · Open-vocab segmentation (avg of 8) | mIoU: 28.4 -> 36.3 | Improved |
| add UniRefiner to SigLIPv2 used as REPA target | SiT-XL/2 + SigLIPv2 · REPA ImageNet 256x256 | FID: 2.21 -> 1.96 | Improved |
| add UniRefiner to SigLIPv2 used as REPA target | SiT-XL/2 + SigLIPv2 · REPA ImageNet 256x256 | sFID: 4.43 -> 4.43 | No clear effect |
| use Gaussian-noise registers instead of learnable registers (ablation) | UniRefiner (SigLIPv2-So400m) · ADE20K linear probing | mIoU: 47.9 -> 49.8 | Improved |
| add uniformity loss L_uni (ablation) | UniRefiner (SigLIPv2-So400m) · ADE20K linear probing | mIoU: 45.8 -> 49.8 | Improved |
| add spurious alignment loss L_spu (ablation) | UniRefiner (SigLIPv2-So400m) · ADE20K linear probing | mIoU: 45.0 -> 49.8 | Improved |
| add FP-GP filter, AH filter and register-based filter (ablation) | UniRefiner (SigLIPv2-So400m) · ADE20K linear probing | mIoU | Improved |
Closest papers by meaning
- Vision Transformers with Self-Distilled Registers (NeurIPS 2025 · similarity 0.89)
- Register and [CLS] tokens induce a decoupling of local and global features in large ViTs (NeurIPS 2025 · similarity 0.89)
- IVC-Prune: Revealing the Implicit Visual Coordinates in LVLMs for Vision Token Pruning (ICLR 2026 · similarity 0.88)
- VITRIX-UniViTAR: Unified Vision Transformer with Native Resolution (NeurIPS 2025 · similarity 0.88)
- Vision Transformers Need Registers (ICLR 2024 · similarity 0.88)
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 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.
