Research map / Vision-language and multimodal / Cross-modal learning
Cross-Modal Center Loss for 3D Cross-Modal Retrieval
Longlong Jing, Elahe Vahdani, Jiaxing Tan, Yingli Tian
Check 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 cross-modal center loss to cross-entropy loss (ablation) | 3D retrieval framework · ModelNet40, nine cross-modal and in-domain retrieval directions | mAP: 75.09 -> 84.87; 75.38 -> 86.7; 69.76 -> 86.11; 75.53 -> 88.83; 75.2 -> 85.66; 69.64 -> 87.58; 66.63 -> 86.89; 69.54 -> 8 | Improved |
| Add mean-square-error loss to cross-entropy and cross-modal center losses (ablation) | 3D retrieval framework · ModelNet40, nine cross-modal and in-domain retrieval directions | mAP: 84.87 -> 86.0; 86.7 -> 87.31; 86.11 -> 86.79; 88.83 -> 88.59; 85.66 -> 85.96; 87.58 -> 87.37; 86.89 -> 87.04; 84.76 -> 8 | Mixed |
| Add mean-square-error loss to cross-entropy loss (ablation) | 3D retrieval framework · ModelNet40, nine cross-modal and in-domain retrieval directions | mAP: 75.09 -> 74.21; 75.38 -> 75.86; 69.76 -> 70.52; 75.53 -> 76.36; 75.2 -> 74.76; 69.64 -> 70.34; 66.63 -> 68.18; 69.54 -> | Mixed |
| Increase batch size from 12 to 96 (ablation) | Cross-modal center loss framework · ModelNet40, nine retrieval directions | mAP: 45.67 -> 90.23; 13.89 -> 89.59; 32.32 -> 89.04; 25.5 -> 88.51; 6.98 -> 88.11; 8.29 -> 87.37; 59.5 -> 87.04; 27.68 -> 87. | Improved |
| Use end-to-end joint training instead of extending DSCMR | 3D cross-modal retrieval · ModelNet40, all nine retrieval directions | mAP: 82.31 -> 90.23; 77.30 -> 89.59; 74.33 -> 89.04; 74.84 -> 88.51; 76.18 -> 88.11; 70.21 -> 87.37; 70.80 -> 87.04; 73.74 -> | Improved |
| Use end-to-end joint training instead of extending DSCMR | 3D cross-modal retrieval · ModelNet10, all nine retrieval directions | mAP: 84.49 -> 91.75; 84.09 -> 91.23; 81.73 -> 91.37; 83.92 -> 90.41; 82.52 -> 89.98; 80.81 -> 90.00; 83.08 -> 90.99; 84.15 -> | Improved |
| Use the proposed framework for in-domain retrieval | 3D object retrieval · ModelNet40 image, point, and mesh domains | mAP: Mesh: 81.9 -> 90.41; Point: 85.0 -> 90.99; Image: 91.1 -> 91.75 | Mixed |
Closest papers by meaning
- RONO: Robust Discriminative Learning With Noisy Labels for 2D-3D Cross-Modal Retrieval (CVPR 2023 · similarity 0.88)
- CrossOver: 3D Scene Cross-Modal Alignment (CVPR 2025 · similarity 0.87)
- DREAM: Decoupled Discriminative Learning with Bigraph-aware Alignment for Semi-supervised 2D-3D Cross-modal Retrieval (AAAI 2025 · similarity 0.86)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.86)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.86)
- Cross-Modal 3D Representation with Multi-View Images and Point Clouds (CVPR 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.
