Pith. sign in

REVIEW 2 cited by

Do Vision Transformers See Like Humans? Evaluating their Perceptual Alignment

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2508.09850 v1 pith:TUMXHEZK submitted 2025-08-13 cs.CV

Do Vision Transformers See Like Humans? Evaluating their Perceptual Alignment

classification cs.CV
keywords alignmentdatasethumanmodelsperceptualaugmentationdataimpact
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Vision Transformers (ViTs) achieve remarkable performance in image recognition tasks, yet their alignment with human perception remains largely unexplored. This study systematically analyzes how model size, dataset size, data augmentation and regularization impact ViT perceptual alignment with human judgments on the TID2013 dataset. Our findings confirm that larger models exhibit lower perceptual alignment, consistent with previous works. Increasing dataset diversity has a minimal impact, but exposing models to the same images more times reduces alignment. Stronger data augmentation and regularization further decrease alignment, especially in models exposed to repeated training cycles. These results highlight a trade-off between model complexity, training strategies, and alignment with human perception, raising important considerations for applications requiring human-like visual understanding.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. TUDSR: Twice Upsampling-Diffusion for Higher Super-Resolution

    cs.CV 2026-06 unverdicted novelty 6.0

    TUDSR applies a twice-upsampling diffusion strategy with chunk-based training to achieve state-of-the-art super-resolution at 1024^2 and 2048^2 resolutions using a one-step GAN on SD2.1-base.

  2. Accuracy Does Not Guarantee Human-Likeness: Cross-Domain Human-Centered Benchmark in Monocular Depth Estimation

    cs.CV 2025-12 conditional novelty 6.0

    Across 69 monocular depth estimators, human-likeness of error patterns peaks near human-level accuracy and declines for the most accurate models: accuracy does not guarantee human-like depth perception.