Research map / Detection and segmentation / Open-set and few-shot detection
MOCHA: Multi-modal Objects-aware Cross-arcHitecture Alignment
Elena Camuffo, Francesco Barbato, Mete Ozay, Simone Milani, Umberto Michieli
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 relational embedding loss L_emb (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 56.6 -> 59.1 | Improved |
| add relational embedding loss L_emb (ablation) | MOCHA (AuXFT) · iCubWorld 1-shot | retrieval accuracy: 59.8 -> 61.4 | Improved |
| use attention in the translator (ablation) | MOCHA (AuXFT) · iCubWorld 5-shot | retrieval accuracy: 73.2 -> 77.0 | Improved |
| keep final ffn layer in translator (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 55.7 -> 59.1 | Improved |
| use non-linear translator instead of linear projection (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 49.3 -> 59.1 | Improved |
| use non-linear translator instead of linear projection (ablation) | MOCHA (AuXFT) · POD 1-shot | mAP50-95: 29.4 -> 36.3 | Improved |
| use all pairs in relational loss instead of Top-K=32 (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 51.1 -> 59.1 | Improved |
| do not enable data augmentation during distillation (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 53.3 -> 59.1 | Improved |
| keep default learning rate instead of doubling encoder lr (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 53.6 -> 59.1 | Improved |
| use PCA dimension 512 instead of 1024 (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 53.8 -> 59.1 | Improved |
| use PCA dimension 512 instead of 256 (ablation) | MOCHA (AuXFT) · PerSeg 1-shot | mAP50-95: 56.4 -> 59.1 | Improved |
| pretrain student on COCO instead of OpenImages only (ablation) | MOCHA · PerSeg 1-shot | mAP50-95: 53.8 -> 55.4 | Improved |
Closest papers by meaning
- Re-Aligning Language to Visual Objects with an Agentic Workflow (ICLR 2025 · similarity 0.87)
- Align-KD: Distilling Cross-Modal Alignment Knowledge for Mobile Vision-Language Large Model Enhancement (CVPR 2025 · similarity 0.87)
- Exploring Region-Word Alignment in Built-in Detector for Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.87)
- Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection (AAAI 2025 · similarity 0.87)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.87)
- Universal Domain Adaptive Object Detection via Dual Probabilistic Alignment (AAAI 2025 · similarity 0.86)
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.
