Research map / Vision-language and multimodal / Cross-modal learning
One Model to Translate Them All: Universal Any-to-Any Translation for Heterogeneous Collaborative Perception
Yang Li, Weize Li, Quan Yuan, Shao Congzhang, Guiyang Luo, Yunqi Ba, Xuanhan Zhu, Xinyuan Ding, Xiaoyuan Fu, Jinglin Li
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 contrastive loss L_IC in stage 1 (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.685/0.575 -> 0.716/0.605 | Improved |
| add modality classification loss L_IS in stage 1 (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.694/0.583 -> 0.716/0.605 | Improved |
| add stage 1 intrinsic-space training (both L_IC and L_IS) (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.662/0.540 -> 0.716/0.605 | Improved |
| add downstream task loss L_task (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.691/0.579 -> 0.716/0.605 | Improved |
| add feature distillation loss L_feat (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.653/0.531 -> 0.716/0.605 | Improved |
| add routing contrastive loss L_ctr (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.688/0.576 -> 0.716/0.605 | Improved |
| add router load-balancing/logit regularizer L_r (ablation) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7: 0.708/0.598 -> 0.716/0.605 | Improved |
| replace Classic MoE (multi-expert forwarding, TopK=3) with parameter-combination single translator | UniTrans · OPV2V-H profiling | GFLOPs: 245.5 -> 109.3 | Improved |
| mapping inferred in intrinsic latent space instead of routing directly on high-dim features (vs Classic MoE) | UniTrans · OPV2V-H | AP@0.5 / AP@0.7 avg: 0.653/0.544 -> 0.716/0.605 | Improved |
| UniTrans vs best baseline on camera ego modalities | UniTrans · OPV2V-H m27 / m30 | AP@0.5/AP@0.7: 0.497/0.243 on m27, 0.406/0.202 on m30 | Improved |
| UniTrans vs NegoCollab in real-world setting | UniTrans · DAIR-V2X | AP@0.5/AP@0.7 avg: 0.509/0.389 -> 0.553/0.421 | Improved |
Closest papers by meaning
- Pragmatic Heterogeneous Collaborative Perception via Generative Communication Mechanism (NeurIPS 2025 · similarity 0.87)
- One is Plenty: A Polymorphic Feature Interpreter for Immutable Heterogeneous Collaborative Perception (CVPR 2025 · similarity 0.87)
- NegoCollab: A Common Representation Negotiation Approach for Heterogeneous Collaborative Perception (NeurIPS 2025 · similarity 0.87)
- An Extensible Framework for Open Heterogeneous Collaborative Perception (ICLR 2024 · similarity 0.86)
- Understanding Task Transfer in Vision-Language Models (CVPR 2026 · similarity 0.86)
- Vision-Language Models Create Cross-Modal Task Representations (ICML 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.
