Research map / Vision-language and multimodal / CLIP and zero-shot
Exploring Cross-Modal Flows for Few-Shot Learning
Ziqi Jiang, Yanghao Wang, 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 FMA multi-step flow-matching alignment on top of CLIP-LoRA (16-shot) | CLIP-LoRA · difficult datasets avg, 16-shot | accuracy: 76.1 -> 77.8 | Improved |
| add FMA on zero-shot CLIP features (16-shot) | CLIP · difficult datasets avg | accuracy: 48.9 -> 68.9 (+20.0) | Improved |
| add FMA on CoOp | CoOp · difficult datasets avg, 16-shot | accuracy: 71.4 -> 74.0 | Improved |
| add FMA on CoCoOp | CoCoOp · difficult datasets avg, 16-shot | accuracy: 64.1 -> 68.5 | Improved |
| add FMA on CLIP-Adapter | CLIP-Adapter · difficult datasets avg, 16-shot | accuracy: 62.4 -> 67.9 | Improved |
| apply FMA to easy datasets rather than difficult ones | CLIP-LoRA · 16-shot easy datasets avg | accuracy: 88.2 -> 88.8 | Improved |
| add FMA when fewer shots (1-shot) on CLIP-LoRA | CLIP-LoRA · 1-shot difficult avg | accuracy: 58.7 -> 59.4 | Improved |
| add FMA trained on ImageNet, evaluated cross-dataset | CoOp · generalization on easy datasets | accuracy: 75.8 -> 75.7 | No clear effect |
| add FMA, harmonic mean of adaptation and generalization | CLIP · difficult datasets | harmonic mean accuracy: 48.9 -> 57.5 | Improved |
| use early stopping solver instead of vanilla flow-matching solver (ablation) | FMA · DTD 16-shot | accuracy | Improved |
| increase inference steps beyond about 8 with early stopping solver (ablation) | FMA · DTD 16-shot | accuracy | Hurt |
| add noise augmentation during velocity training (ablation) | FMA on CLIP-LoRA · difficult datasets | accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- PLOT++ — PLOT: Prompt Learning with Optimal Transport for Vision-Language Models (ICLR 2023) · better on 4 of 4 shared results
- CLIP-Adapter — CLIP-Adapter: Better Vision-Language Models with Feature Adapters (IJCV 2023) · better on 1 of 1 shared result
- TIP-Adapter — Tip-Adapter: Training-Free Adaption of CLIP for Few-Shot Classification (ECCV 2022) · better on 1 of 1 shared result
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
- Direct Product Flow Matching: Decoupling Radial and Angular Dynamics for Few-Shot Adaptation (NeurIPS 2026 · similarity 0.90)
- Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution (CVPR 2025 · similarity 0.89)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.88)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.87)
- Linguistic-Aware Patch Slimming Framework for Fine-grained Cross-Modal Alignment (CVPR 2024 · similarity 0.87)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.87)
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.
