Research map / Vision-language and multimodal / Cross-modal learning
RoRE: Rotary Ray Embedding for Generalised Multi-Modal Scene Understanding
Ryan Griffiths, Donald G. Dansereau
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 ray-based relative rotary embedding on top of absolute embedding (ablation) | LVSM · RealEstate10K | PSNR: 26.18 -> 26.56 | Improved |
| add asymmetric positions (negated shifted components) (ablation) | RoRE · RealEstate10K | PSNR: 26.56 -> 26.65 | Improved |
| use learnt rotation frequencies instead of handcrafted (ablation) | RoRE · RealEstate10K | PSNR: 26.65 -> 26.65 | No clear effect |
| remove absolute embedding from RoRE (ablation) | RoRE · RealEstate10K | PSNR: 26.65 -> 26.65 | No clear effect |
| RoRE versus LVSM when focal length varies without retraining | RoRE · RealEstate10K, varied focal length | PSNR: 21.95 -> 22.66 | Improved |
| RoRE versus GTA/PRoPE under varied focal length | RoRE · RealEstate10K, varied focal length | PSNR: 14.81 -> 22.66 vs GTA | Improved |
| RoRE versus LVSM on distorted inputs | RoRE · Distorted RealEstate10K | PSNR: 21.99 -> 23.96 | Improved |
| RoRE versus LVSM on native fisheye | RoRE · Fisheye FIORD | PSNR: 22.52 -> 23.55 | Improved |
| RoRE versus PRoPE/GTA on standard perspective data | RoRE · RealEstate10K | PSNR: 26.81 (PRoPE) vs 26.65 (RoRE) | No clear effect |
| RoRE versus LVSM on in-domain NVS | RoRE · RealEstate10K | PSNR: 26.18 -> 26.65 vs LVSM; DL3DV 19.48 -> 19.77 | Mixed |
| double network depth from 6 to 12 layers (ablation) | RoRE · RealEstate10K | PSNR: 26.65 -> 28.79 | Improved |
| increase input mask ratio from 0% to 50% at inference (ablation) | RoRE multimodal · MultiModalBlender | RGB PSNR: 21.494 -> 18.287 | Hurt |
Closest papers by meaning
- URoPE: Universal Relative Position Embedding across Geometric Spaces (ECCV 2026 · similarity 0.86)
- ComRoPE: Scalable and Robust Rotary Position Embedding Parameterized by Trainable Commuting Angle Matrices (CVPR 2025 · similarity 0.85)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.84)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 2025 · similarity 0.84)
- LieRE: Lie Rotational Positional Encodings (ICML 2025 · similarity 0.83)
- OpenHype: Hyperbolic Embeddings for Hierarchical Open-Vocabulary Radiance Fields (NeurIPS 2025 · similarity 0.83)
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.
