Research map / Vision-language and multimodal / CLIP and zero-shot
FlowComposer: Composable Flows for Compositional Zero-Shot Learning
Zhenqi He, Lin Li, Long Chen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add FlowComposer to Troika | Troika · UT-Zappos (closed-world) | AUC: 41.8 -> 46.8 | Improved |
| add FlowComposer to CSP | CSP · C-GQA (closed-world) | AUC: 5.8 -> 7.7 | Improved |
| add FlowComposer to Troika | Troika · MIT-States (open-world) | AUC: 6.8 -> 7.5 | Improved |
| add primitive flows (ablation) | Troika · UT-Zappos (closed-world) | AUC: 41.8 -> 43.0 | Improved |
| add Composer on top of flows (ablation) | Troika + flows · UT-Zappos (closed-world) | AUC: 43.0 -> 45.8 | Improved |
| add leakage-guided augmentation on top of flows and Composer (ablation) | Troika + flows + Composer · UT-Zappos (closed-world) | AUC: 45.8 -> 46.8 | Improved |
| add leakage-guided augmentation on top of flows and Composer (ablation) | Troika + flows + Composer · MIT-States (closed-world) | AUC: 23.3 -> 23.5 | Improved |
| use attribute and object flows instead of a single composition flow (ablation) | Troika · UT-Zappos (closed-world) | AUC: 37.2 -> 46.8 | Improved |
| use Composer instead of an extra composition flow (c+a+o) (ablation) | Troika+FlowComposer · MIT-States (closed-world) | AUC: 20.5 -> 23.5 | Improved |
| use Composer coefficients instead of a direct velocity predictor (ablation) | Troika+FlowComposer · UT-Zappos (closed-world) | AUC: 45.2 -> 46.8 | Improved |
| use Composer instead of predictor (ablation) | Troika+FlowComposer · MIT-States (closed-world) | AUC: 23.3 -> 23.5 | Improved |
| flow matching instead of parameter-matched regressor (ablation) | Troika+FlowComposer · C-GQA (closed-world) | AUC: 12.9 -> 15.9 | Improved |
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.
- Troika — Troika: Multi-Path Cross-Modal Traction for Compositional Zero-Shot Learning (CVPR 2024) · better on 6 of 6 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
- Disentangling Before Composing: Learning Invariant Disentangled Features for Compositional Zero-Shot Learning (TPAMI 2024 · similarity 0.92)
- Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning (CVPR 2024 · similarity 0.89)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- LOGICZSL: Exploring Logic-induced Representation for Compositional Zero-shot Learning (CVPR 2025 · similarity 0.89)
- A Dynamic Learning Method towards Realistic Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.89)
- Beyond Image Classification: A Video Benchmark and Dual-Branch Hybrid Discrimination Framework for Compositional Zero-Shot Learning (CVPR 2025 · similarity 0.88)
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.
