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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Sparse Alignment (max over patch tokens) training (ablation) | CLIP ViT-L/14 baseline · UT-Zappos closed-world | AUC: 37.5 -> 44.4 | Improved |
| add Sparse Alignment (ablation) | CLIP ViT-L/14 baseline · C-GQA closed-world | AUC: 14.3 -> 15.2 | Improved |
| add VAC with KL distillation from SA (ablation) | SA model · UT-Zappos closed-world | AUC: 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-world | AUC: 48.6 -> 49.3 | Improved |
| dynamically update memory bank with test samples including unseen compositions (ablation) | SA + VAC + Memory Bank · UT-Zappos closed-world | AUC: 49.3 -> 50.0 | Improved |
| use max over patches instead of mean pooling (ablation) | Sparse Alignment · C-GQA | AUC: 13.9 -> 15.2 | Improved |
| use max instead of learned attention or linear patch weights (ablation) | Sparse Alignment · C-GQA | AUC: 13.5 (Attention), 14.5 (Linear) -> 15.2 | Improved |
| add Sparse Alignment to Troika | Troika · C-GQA | AUC: 12.4 -> 13.1 | Improved |
| add Sparse Alignment to Troika | Troika · UT-Zappos | AUC: 41.7 -> 43.1 | Improved |
| randomly drop input patch tokens (naive visual reduction) (ablation) | CLIP LoRA finetune · C-GQA | AUC: 14.3 -> 14.7 | Improved |
| high random patch drop rates (ablation) | CLIP LoRA finetune · C-GQA | AUC | Hurt |
| add SA, then VAC (ablation) | Baseline · UT-Zappos | relative modality gap (RMG): 0.1825 -> 0.1271 -> 0.1072 | Improved |
Closest papers by meaning
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.91)
- A Dynamic Learning Method towards Realistic Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.91)
- Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.90)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- TOMCAT: Test-time Comprehensive Knowledge Accumulation for Compositional Zero-Shot Learning (NeurIPS 2025 · similarity 0.89)
- Troika: Multi-Path Cross-Modal Traction for Compositional Zero-Shot Learning (CVPR 2024 · similarity 0.89)
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.
