Spiral RoPE: Rotate Your Rotary Positional Embeddings in the 2D Plane
This addresses a fundamental limitation in positional encoding for vision transformers, improving modeling of oblique spatial relationships in natural images, though it appears incremental as an extension of existing RoPE.
The paper tackles the limitation of standard 2D Rotary Position Embedding (RoPE) in vision transformers, which restricts positional encoding to axis-aligned directions, by proposing Spiral RoPE to enable multi-directional encoding, resulting in consistent performance improvements across classification, segmentation, and generation tasks.
Rotary Position Embedding (RoPE) is the de facto positional encoding in large language models due to its ability to encode relative positions and support length extrapolation. When adapted to vision transformers, the standard axial formulation decomposes two-dimensional spatial positions into horizontal and vertical components, implicitly restricting positional encoding to axis-aligned directions. We identify this directional constraint as a fundamental limitation of the standard axial 2D RoPE, which hinders the modeling of oblique spatial relationships that naturally exist in natural images. To overcome this limitation, we propose Spiral RoPE, a simple yet effective extension that enables multi-directional positional encoding by partitioning embedding channels into multiple groups associated with uniformly distributed directions. Each group is rotated according to the projection of the patch position onto its corresponding direction, allowing spatial relationships to be encoded beyond the horizontal and vertical axes. Across a wide range of vision tasks including classification, segmentation, and generation, Spiral RoPE consistently improves performance. Qualitative analysis of attention maps further show that Spiral RoPE exhibits more concentrated activations on semantically relevant objects and better respects local object boundaries, highlighting the importance of multi-directional positional encoding in vision transformers.