Phase Marginalization for Patch-Grid Instability in Vision Transformers

Hugging Face Daily Papers Papers

Summary

Phase Marginalization is a post-hoc method that addresses phase-dependent instability in Vision Transformers by evaluating structured patch-grid phases and aggregating outputs. It improves segmentation, depth, and local matching over the canonical baseline with minimal extra cost.

Vision Transformers operate on fixed patch grids, which can introduce phase-dependent instability for dense prediction: changing the patch partition can change the token evidence available to a pixel, especially near boundaries. We formalize patch-grid phase as a nuisance variable and propose Phase Marginalization, a post-hoc marginalization method that evaluates structured patch-grid phases, inverse-aligns dense outputs, and aggregates them in the original image coordinate system. The central variant, Uniform Phase Marginalization with K = 4, is training-free and improves over the canonical K = 1 baseline across measured segmentation, depth, and local matching settings. In a controlled Cityscapes experiment, Uniform Phase Marginalization provides a modest compute-matched advantage over generic shift-based four-forward test-time augmentation (TTA) (+0.31 mean Intersection-over-Union over the strongest tested generic row). A scaling study further shows that K = 4 is a practical cost-accuracy trade-off: K = 8 is essentially unchanged and K = 16 adds little accuracy at much higher latency. These results position patch-grid phase as a measurable nuisance variable and Phase Marginalization as a simple diagnostic and post-hoc marginalization baseline for dense ViT prediction.
Original Article
View Cached Full Text

Cached at: 06/10/26, 12:13 AM

Paper page - Phase Marginalization for Patch-Grid Instability in Vision Transformers

Source: https://huggingface.co/papers/2606.08132

Abstract

Phase Marginalization is a post-hoc method that addresses phase-dependent instability in Vision Transformers by evaluating structured patch-grid phases and aggregating outputs in the original image coordinate system.

Vision Transformersoperate on fixedpatch grids, which can introducephase-dependent instabilityfordense prediction: changing the patch partition can change the token evidence available to a pixel, especially near boundaries. We formalizepatch-grid phaseas a nuisance variable and proposePhase Marginalization, a post-hoc marginalization method that evaluatesstructured patch-grid phases,inverse-alignsdense outputs, and aggregates them in the originalimage coordinate system. The central variant, UniformPhase Marginalizationwith K = 4, is training-free and improves over the canonical K = 1 baseline across measuredsegmentation, depth, andlocal matchingsettings. In a controlledCityscapesexperiment, UniformPhase Marginalizationprovides a modest compute-matched advantage over generic shift-based four-forwardtest-time augmentation(TTA) (+0.31 meanIntersection-over-Unionover the strongest tested generic row). A scaling study further shows that K = 4 is a practical cost-accuracy trade-off: K = 8 is essentially unchanged and K = 16 adds little accuracy at much higher latency. These results positionpatch-grid phaseas a measurable nuisance variable andPhase Marginalizationas a simple diagnostic and post-hoc marginalization baseline for dense ViT prediction.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2606\.08132

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.08132 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.08132 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.08132 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Variable-Width Transformers

Hugging Face Daily Papers

Proposes a nonuniform width allocation transformer (hourglass shape) that outperforms uniform baselines in language modeling, reducing FLOPs and KV cache size.

PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation

Hugging Face Daily Papers

PointDiT presents a minimalist pixel-space diffusion transformer using a plain ViT architecture for monocular geometry estimation, outperforming complex latent-based models while maintaining simplicity and robustness in ambiguous regions.