Research map / Vision-language and multimodal / CLIP and zero-shot
RMAdapter: Reconstruction-based Multi-Modal Adapter for Vision-Language Models
Xiang Lin, Weixin Li, Shu Guo, Lihong Wang, Di Huang
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 reconstruction branch and consistency constraint to MMA (ablation) | MMA · Base-to-Novel, avg of 11 datasets | HM: 79.87 -> 80.62 | Improved |
| add consistency constraint alone to MMA (ablation) | MMA · Base-to-Novel, avg of 11 datasets | HM: base 83.20 -> 83.86; novel 76.80 -> 76.31 | Mixed |
| add text reconstruction on top of constraints (ablation) | MMA + constraints · Base-to-Novel, avg of 11 datasets | HM: 79.91 -> 80.51 | Improved |
| add visual reconstruction on top of constraints (ablation) | MMA + constraints · Base-to-Novel, avg of 11 datasets | HM: 79.91 -> 80.39 | Improved |
| share down-projection instead of independent projections (ablation) | RMAdapter · Base-to-Novel, avg of 11 datasets | HM: 80.52 -> 80.62 | Improved |
| share down-projection instead of sharing up-projection (ablation) | RMAdapter · Base-to-Novel, avg of 11 datasets | HM: 80.30 -> 80.62 | Improved |
| use two-layer instead of single-layer up-projection (ablation) | RMAdapter · Base-to-Novel, avg of 11 datasets | HM: 80.57 -> 80.62 | Improved |
| use three-layer instead of two-layer up-projection (ablation) | RMAdapter · Base-to-Novel, avg of 11 datasets | HM: 80.62 -> 80.08 | Hurt |
| use RMAdapter instead of MMA | RMAdapter · Cross-dataset (ImageNet-trained, 10 targets zero-shot) | avg accuracy: 66.61 -> 67.56 | Improved |
| use RMAdapter instead of CoPrompt | RMAdapter · Base-to-Novel HM, EuroSAT | HM: 85.84 vs 84.69 | Hurt |
| use RMAdapter instead of CoPrompt | RMAdapter · Domain generalization (ImageNet variants avg) | avg accuracy: 60.42 -> 60.71 | Improved |
Closest papers by meaning
- MMA: Multi-Modal Adapter for Vision-Language Models (CVPR 2024 · similarity 0.92)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.91)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.90)
- UniAdapter: Unified Parameter-Efficient Transfer Learning for Cross-modal Modeling (ICLR 2024 · similarity 0.90)
- A Wander Through the Multimodal Landscape: Efficient Transfer Learning via Low-rank Sequence Multimodal Adapter (AAAI 2025 · similarity 0.90)
- IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities (AAAI 2025 · 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.
