Research map / Vision-language and multimodal / CLIP and zero-shot
Multi-Modal Few-Shot Object Detection with Meta-Learning-Based Cross-Modal Prompting
Guangxing Han, Long Chen, Jiawei Ma, Shiyuan Huang, Rama Chellappa, Shih‐Fu Chang
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 the language (multi-modal) branch to the vision-only model (ablation) | MM-FSOD · MSCOCO 10-shot meta-testing | AP: 9.7 -> 10.8 | Improved |
| use generated soft prompts (teacher) instead of shared soft prompt across classes (ablation) | text classifier · MSCOCO 10-shot meta-testing | AP: 1.6 -> 4.7 | Improved |
| distill student without class names from the teacher (ablation) | prompt generator · MSCOCO 10-shot meta-testing | AP: 4.7 -> 4.8 | No clear effect |
| use context 'a photo of [class_name]' instead of bare class name (ablation) | hard prompt · MSCOCO meta-testing | AP: 2.2 -> 3.0 | Improved |
| increase learnable prompt tokens from 2 to 8 (ablation) | soft prompt · MSCOCO 10-shot meta-testing | AP: 4.0 -> 4.8 | Improved |
| increase prompt tokens from 8 to 16 (ablation) | soft prompt · MSCOCO 10-shot meta-testing | AP: 4.8 -> 4.6 | No clear effect |
| place prompt tokens before the class name instead of after (ablation) | teacher soft prompt · MSCOCO 10-shot meta-testing | AP: 4.6 -> 4.8 | Improved |
| use two MLP layers instead of one in generator (ablation) | soft prompt generator G · MSCOCO 10-shot meta-testing | AP: 4.8 -> 4.4 | Hurt |
| use post-model transformer generator instead of one-layer MLP (ablation) | soft prompt generator G · MSCOCO 10-shot meta-testing | AP: 4.8 -> 2.6 | Hurt |
| use addition instead of concatenation for multi-modal fusion (ablation) | MPG module · MSCOCO 10-shot meta-testing | AP: 9.0 -> 10.8 | Improved |
| apply MPG to RCNN classification only versus vision-only (ablation) | detection model · MSCOCO 10-shot meta-testing | AP: 9.7 -> 10.5 | Improved |
| also apply MPG to the RPN (ablation) | detection model · MSCOCO 10-shot meta-testing | AP: 10.5 -> 10.8 | Improved |
Closest papers by meaning
- Cross-domain Few-shot Object Detection with Multi-modal Textual Enrichment (IJCV 2026 · similarity 0.92)
- DetPO: In-Context Learning with Multi-Modal LLMs for Few-Shot Object Detection (ECCV 2026 · similarity 0.91)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.90)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · 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.
