AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Bridging the Modality Gap in Compositional Zero-Shot Learning via Sparse Alignment and Unimodal Memory Bank

Yang Zhang, Zhixiang Chi, Xudong Yan, Yang Wang, Songhe Feng

CVPR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Attribute-based zero-shot learning

Read the paper

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 Sparse Alignment (max over patch tokens) training (ablation)CLIP ViT-L/14 baseline · UT-Zappos closed-worldAUC: 37.5 -> 44.4Improved
add Sparse Alignment (ablation)CLIP ViT-L/14 baseline · C-GQA closed-worldAUC: 14.3 -> 15.2Improved
add VAC with KL distillation from SA (ablation)SA model · UT-Zappos closed-worldAUC: 44.4 -> 48.6 (with 46.2 for cross-entropy only)Improved
add static memory bank for seen compositions (ablation)SA + VAC · UT-Zappos closed-worldAUC: 48.6 -> 49.3Improved
dynamically update memory bank with test samples including unseen compositions (ablation)SA + VAC + Memory Bank · UT-Zappos closed-worldAUC: 49.3 -> 50.0Improved
use max over patches instead of mean pooling (ablation)Sparse Alignment · C-GQAAUC: 13.9 -> 15.2Improved
use max instead of learned attention or linear patch weights (ablation)Sparse Alignment · C-GQAAUC: 13.5 (Attention), 14.5 (Linear) -> 15.2Improved
add Sparse Alignment to TroikaTroika · C-GQAAUC: 12.4 -> 13.1Improved
add Sparse Alignment to TroikaTroika · UT-ZapposAUC: 41.7 -> 43.1Improved
randomly drop input patch tokens (naive visual reduction) (ablation)CLIP LoRA finetune · C-GQAAUC: 14.3 -> 14.7Improved
high random patch drop rates (ablation)CLIP LoRA finetune · C-GQAAUCHurt
add SA, then VAC (ablation)Baseline · UT-Zapposrelative modality gap (RMG): 0.1825 -> 0.1271 -> 0.1072Improved

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