AI Paper Atlas

Research map / Label-efficient and robust learning / Noisy labels

TANGO: Text-Anchored Guided Optimization for Robust Fine-tuning Vision-Language Models under Label Noise

Tengfei Ma, Weiran Pan, Wei Wei

CVPR 2026 · Label-efficient and robust learning › Noisy labels · approach: Noisy cross-modal hashing and multi-label

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 Text-Anchored Classifier and anchor-guided refinement (full TANGO vs baseline) (ablation)TANGO · CIFAR-100N R40%accuracy (%): 81.97 -> 83.83Improved
add semantic verification to visual-only refinement (ablation)TANGO · CIFAR-100N R40%accuracy (%): 83.47 -> 83.83Improved
add visual neighborhood information to semantic-only refinement (ablation)TANGO · CIFAR-100N R40%accuracy (%): 83.58 -> 83.83Improved
use semantic-only instead of visual-only refinement (ablation)TANGO · CIFAR-100 Sym. 40%accuracy (%): 89.43 -> 89.79Improved
make anchors trainable instead of fixed (ablation)TANGO · CIFAR-100N R40%accuracy (%): 83.83 -> 84.01No clear effect
use simpler LLM-generated prompts instead of curated CuPL prompts (ablation)TANGO · CIFAR-100 Sym. 40%accuracy (%): 89.82 -> 89.22Hurt
increase anchors per class from 5 to 10 (ablation)TANGO · CIFAR-100Naccuracy (%): +1.54%Improved
increase anchors per class beyond 10 (ablation)TANGO · CIFAR-100Naccuracy (%)No clear effect
vary semantic weight beta from 0 to 1 (ablation)TANGO · CIFAR-100Naccuracy (%): peak 83.83 at beta=0.5No clear effect
replace 90% of anchors with random noise (ablation)TANGO · CIFAR-100Naccuracy (%): 83.83 -> 82.51Hurt
replace all anchors with random noise (ablation)TANGO · CIFAR-100Naccuracy (%): 83.83 -> 80.55Hurt
use TANGO instead of LSL on SigLIP backboneTANGO · CIFAR-100Naccuracy (%): 73.26 -> 85.31Improved

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.

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