Research map / Vision-language and multimodal / Human and embodied modeling
Learning multiple visual domains with residual adapters
Read the paperCheck 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 |
|---|---|---|---|
| Add residual adapter modules instead of using only the frozen feature extractor | ResNet28 · Aircraft | top-1 accuracy (%): 23.31% → 56.68% | Helped |
| Use higher adapter weight decay for small target datasets (ablation) | Res. adapt. · Flowers102 | top-1 accuracy (%): 66.24 -> 81.67 | Helped |
| Use higher adapter weight decay (ablation) | Res. adapt. · DTD | top-1 accuracy (%): 50.85 -> 57.13 | Helped |
| End-to-end fine-tune all network parameters across the ten tasks (ablation) | Res. adapt. · Visual Decathlon | score: 2118 -> 2643 | Helped |
| Predict the input domain with an auxiliary network rather than provide it (ablation) | Res. adapt. · Visual Decathlon | score: 2621 -> 2503 | Hurt |
| Remove residual connections from the adapter modules (ablation) | Res. adapt. · Ten visual domains | mean top-1 accuracy (%): 73.9 -> 70.3 | Hurt |
| Fine-tune only the last two convolutional layers instead of using residual adapters (ablation) | ResNet28 · Ten visual domains | mean top-1 accuracy (%): 73.9 -> 64.7 | Hurt |
| Train separate models from scratch with the added adapter parameters (ablation) | Res. adapt. · Ten visual domains | mean top-1 accuracy (%): 70.32 vs 70.32 | No clear effect |
| Use a larger-capacity residual-adapter network (ablation) | Res. adapt. · Visual Decathlon | score: 2621 -> 3131 | Helped |
Closest papers by meaning
- Visual Representation Learning over Latent Domains (ICLR 2022 · similarity 0.90)
- Depthwise Convolution Is All You Need for Learning Multiple Visual Domains (AAAI 2019 · similarity 0.89)
- AdaGraph: Unifying Predictive and Continuous Domain Adaptation Through Graphs (CVPR 2019 · similarity 0.88)
- Diversify and Match: A Domain Adaptive Representation Learning Paradigm for Object Detection (CVPR 2019 · similarity 0.87)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.86)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.85)
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.