Research map / Vision-language and multimodal / CLIP and zero-shot
VL-Adapter: Parameter-Efficient Transfer Learning for Vision-and-Language Tasks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use one shared adapter across tasks instead of full fine-tuning | CLIP-BART · NLVR² test-P | accuracy (%): 73.0 -> 74.2 | Helped |
| Use one shared adapter across tasks instead of full fine-tuning | CLIP-BART · COCO Caption Karpathy test | CIDEr: 112.9 -> 114.9 | Helped |
| Use one shared adapter across tasks instead of full fine-tuning | CLIP-BART · VQAv2 Karpathy test | accuracy (%): 67.6 -> 65.9 | Hurt |
| Use one shared adapter across tasks instead of full fine-tuning | CLIP-BART · How2QA test-public | accuracy (%): 73.9 -> 73.9 | No clear effect |
| Use one shared adapter across tasks instead of full fine-tuning | CLIP-BART · TVQA, How2QA, TVC, and YC2C | average score: 87.4 -> 87.4 | No clear effect |
| Add adapter modules alongside projection and layer norm updates (ablation) | CLIP-BART + Single Adapter · VQAv2 | accuracy: 49.5 -> 65.9 | Helped |
| Add layer normalization updates (ablation) | CLIP-BART parameter-efficient tuning · VQAv2 | accuracy: 32.2 -> 49.5 | Helped |
| Add task-specific prompts (ablation) | CLIP-BART full fine-tuning · VQAv2 | accuracy: 66.7 -> 67.6 | Helped |
| Add task-specific prompts (ablation) | CLIP-BART + Single Adapter · VQAv2 | accuracy: 65.1 -> 65.9 | Helped |
| Reduce adapter hidden dimension from 96 to 48 (ablation) | CLIP-BART + Single Adapter · VQAv2 | accuracy: 65.9 -> 64.7 | Hurt |
| Reduce adapter hidden dimension from 96 to 24 (ablation) | CLIP-BART + Single Adapter · VQAv2 | accuracy: 65.9 -> 63.5 | Hurt |
| Use V&L-pretrained weights before downstream adaptation (ablation) | CLIP-BART + Single Adapter · VQAv2 | accuracy: 65.9 -> 69.4 | Helped |
Closest papers by meaning
- p-Laplacian Adaptation for Generative Pre-trained Vision-Language Models (AAAI 2024 · similarity 0.91)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.91)
- READ-PVLA: Recurrent Adapter with Partial Video-Language Alignment for Parameter-Efficient Transfer Learning in Low-Resource Video-Language Modeling (AAAI 2024 · similarity 0.90)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.90)
- MMA: Multi-Modal Adapter for Vision-Language Models (CVPR 2024 · similarity 0.90)
- Vision-Language Model Selection and Reuse for Downstream Adaptation (ICML 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.