HRoPE: Homography Relative Positional Encoding
We introduce HRoPE, which represents the local geometry of each patch as a slanted 3D plane, following the slanted support-window assumption of classical multi-view stereo.
ProofPaper ↗
Key points
- Geometric Relative Positional Encoding has become the primary mechanism for injecting spatial inductive biases into multi-view transformers.
- Existing designs condition attention on camera-level frustums or ray-marched depth, associating cross-view tokens through the sensor geometry.
- In this paper, we advocate a shift toward surface-aware positional encoding that explicitly grounds patch tokens in the geometry of physical surfaces.
- Extensive experiments on RealEstate10K, CO3Dv2, and DL3DV show that our best variant of HRoPE consistently outperforms the evaluated geometric positional encodings for novel view synthesis.
Sources (2)
- [1]HRoPE: Homography Relative Positional EncodingHugging Face Daily Papers · Oct 20, 12:00 AM
We introduce HRoPE, which represents the local geometry of each patch as a slanted 3D plane, following the slanted support-window assumption of classical multi-view stereo.
Geometric Relative Positional Encoding has become the primary mechanism for injecting spatial inductive biases into multi-view transformers.
- [2]HRoPE: Homography Relative Positional EncodingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 02:36 PM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026PnP-Transformer: A Neural Pose Solver for 6D Object Pose Estimation
- Oct 19, 2026TorchNeuron: running HF models on AWS Trainium is easier than ever
- Oct 19, 2026TokenGate: Adapting LLMs for Fast and Memory-Efficient Prefill with Learned Token Pruning
- Oct 18, 2026ReDT: Resolution-Decoupled Visual Geometry Transformer
- Oct 18, 2026The Explainability Illusion: Must ViTs Exhibit Patch-Decoding Interpretability?
- Oct 18, 2026ClinSDT: LLM-Encoded Clinical Semantic Guidance for Sepsis Treatment via Decision Transformer