AI Paper Atlas

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

CVPR 2026 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add UniRefinerEVA-CLIP 8B/14 · ADE20K linear probingmIoU: 42.5 -> 51.9Improved
add UniRefinerDINOv2 G/14 · ADE20K linear probingmIoU: 49.1 -> 50.6Improved
add UniRefinerSigLIPv2 So/16 · ADE20K linear probingmIoU: 45.6 -> 49.8Improved
add UniRefinerEVA-CLIP 8B/14 · NYUv2 depth linear probingdelta1: 77.9 -> 90.9Improved
add UniRefinerEVA-CLIP 8B/14 · Open-vocab segmentation CityscapesmIoU: 11.4 -> 30.1Improved
add UniRefinerSigLIPv2 So/16 · Open-vocab segmentation (avg of 8)mIoU: 28.4 -> 36.3Improved
add UniRefiner to SigLIPv2 used as REPA targetSiT-XL/2 + SigLIPv2 · REPA ImageNet 256x256FID: 2.21 -> 1.96Improved
add UniRefiner to SigLIPv2 used as REPA targetSiT-XL/2 + SigLIPv2 · REPA ImageNet 256x256sFID: 4.43 -> 4.43No clear effect
use Gaussian-noise registers instead of learnable registers (ablation)UniRefiner (SigLIPv2-So400m) · ADE20K linear probingmIoU: 47.9 -> 49.8Improved
add uniformity loss L_uni (ablation)UniRefiner (SigLIPv2-So400m) · ADE20K linear probingmIoU: 45.8 -> 49.8Improved
add spurious alignment loss L_spu (ablation)UniRefiner (SigLIPv2-So400m) · ADE20K linear probingmIoU: 45.0 -> 49.8Improved
add FP-GP filter, AH filter and register-based filter (ablation)UniRefiner (SigLIPv2-So400m) · ADE20K linear probingmIoUImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy