SpiralFovea is a parameter-free input-adaptive tokenization method that replaces fixed ViT grids with entropy-driven multi-scale spiral patches, delivering 1.7-2.1 pp accuracy gains and 60% fewer tokens on fine-grained benchmarks.
Indian Conference on Computer Vision, Graphics and Image Processing (
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SpiralFovea: Input-Adaptive Foveated Tokenization as a Third Lever of Resource-Adaptive Inference
SpiralFovea is a parameter-free input-adaptive tokenization method that replaces fixed ViT grids with entropy-driven multi-scale spiral patches, delivering 1.7-2.1 pp accuracy gains and 60% fewer tokens on fine-grained benchmarks.