AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

STD-Former: Image-Conditioned Texture Dictionary Encoding with Sparse Topological Supervision for Texture Recognition

Bo Peng, Ke Xu, Yurui Pan

ICML 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training

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
replace naive pooling with the TDE dictionary encoder (ablation)DINOv2 end-to-end baseline · DTDaccuracy (%): 81.2 -> 82.1Improved
add self-supervised coverage loss (single stage) (ablation)TDE · DTDaccuracy (%): 82.1 -> 83.2Improved
use decoupled two-stage training (coverage pre-train then classifier) (ablation)STD-Former · DTDaccuracy (%): 83.2 -> 84.5Improved
add sparse topological loss (0D/MST) in stage 2 (ablation)STD-Former (DINOv2) · DTDaccuracy (%): 84.5 -> 86.1Improved
add sparse topological loss (0D/MST) in stage 2 (ablation)STD-Former (DINOv2) · GTOSaccuracy (%): 90.7 -> 91.9Improved
use both intra and inter topo terms rather than one (ablation)STD-Former+ · DTDaccuracy (%): 85.1 (intra only) / 85.0 (inter only) -> 86.1Improved
replace mean pooling of textons with attention pooling, Set Transformer PMA or concat+MLP (ablation)STD-Former · DTDaccuracy (%): 84.5 -> 84.6 to 84.8No clear effect
switch backbone from ResNet to DINOv2STD-Former+ · DTDaccuracy (%): 81.7 -> 86.1Improved
add dictionary-based TDE instead of per-image KMeans prototypesSTD-Former · texture benchmarksaccuracyImproved
use image-conditioned textons instead of static training-set dictionarySTD-Former · label-preserving transforms (robustness)accuracy dropImproved

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