AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Co-Me: Confidence Guided Token Merging for Visual Geometric Transformers

Yutian Chen, Yuheng Qiu, Ruogu Li, Jay Patrikar, Sebastian Scherer

CVPR 2026 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 4 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
apply Co-Me token merging to VGGTVGGT* · RealEstate-10K pose (128 frames)speedup: 7.97x -> 16.2xImproved
apply Co-Me token merging to VGGTVGGT* · RealEstate-10K pose (128 frames)AUC translation @10cm: 0.902 -> 0.869Hurt
apply Co-Me to VGGTVGGT · DTU-MVS depth (32 frames)L1 depth error (cm): 0.89 -> 1.10Hurt
apply Co-Me to MapAnythingMapAnything · DTU-MVS depth (32 frames)L1 depth error (cm): 4.59 -> 6.80Hurt
apply Co-Me to DA3-Giant on single-frame NYUd-v2DA3-Giant · NYUd-v2 depth (1 frame)L1 depth error (m): 0.272 -> 0.463Hurt
apply Co-Me to pi3 for ETH3D point cloudpi3 · ETH3D point cloud (16 frames)completeness (cm): 20.8 -> 18.9Improved
merge by predicted confidence rather than token similarity (ablation)Co-Me · DTU-MVS depth (32 frames)speed-accuracy tradeoffImproved
merge low-confidence tokens instead of pick-one or drop-all (ablation)Co-Me · DTU-MVS depth (32 frames)performance degradation: over 10x smaller degradationImproved
add attention bias correction (log n) (ablation)Co-Me · DTU depth and RE10K poseL1 error / pose accuracy: 4x lower delta L1 on DTUImproved
replace baseline VGGT attention with FlashAttention (VGGT*)VGGT · DTU-MVS depth (32 frames)speedup: 1.00x -> 4.81xImproved
use Co-Me on single-frame inputVGGT · ETH3D (1 frame)speedup: 1.00x -> 1.13xNo clear effect
increase input sequence lengthCo-Me + VGGT · speedup vs VGGTspeedup: 4.8x at 32 frames to 21.5x at 512Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most 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

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