Pith. sign in

REVIEW 5 cited by

Discrete Representations Strengthen Vision Transformer Robustness

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

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 2111.10493 v2 pith:U5G67EGI submitted 2021-11-20 cs.CV

Chengzhi Mao , Lu Jiang , Mostafa Dehghani , Carl Vondrick , Rahul Sukthankar , Irfan Essa This is my paper
classification cs.CV
keywords discretevitsarchitectureimagenetinformationrobustnesstokensadding
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Vision Transformer (ViT) is emerging as the state-of-the-art architecture for image recognition. While recent studies suggest that ViTs are more robust than their convolutional counterparts, our experiments find that ViTs trained on ImageNet are overly reliant on local textures and fail to make adequate use of shape information. ViTs thus have difficulties generalizing to out-of-distribution, real-world data. To address this deficiency, we present a simple and effective architecture modification to ViT's input layer by adding discrete tokens produced by a vector-quantized encoder. Different from the standard continuous pixel tokens, discrete tokens are invariant under small perturbations and contain less information individually, which promote ViTs to learn global information that is invariant. Experimental results demonstrate that adding discrete representation on four architecture variants strengthens ViT robustness by up to 12% across seven ImageNet robustness benchmarks while maintaining the performance on ImageNet.

Discussion (0). Sign in to comment.

Forward citations

Cited by 5 Pith papers

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

  1. ArcVQ-VAE: A Spherical Vector Quantization Framework with ArcCosine Additive Margin

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    ArcVQ-VAE constrains VQ-VAE codebook vectors inside a time-dependent ball and adds angular margin loss to increase separability and codebook utilization.

  2. EEGDancer: Dynamic Emotion Latent Space Masked Modeling with Reinforcement Learning for EEG Continuous Emotion Prediction

    cs.HC 2026-06 unverdicted novelty 5.0 of 10

    EEGDancer integrates VQ-VAE latent space learning, masked Transformer modeling, and SAC reinforcement learning to improve continuous EEG emotion prediction over prior methods on SEED datasets.

  3. ArcVQ-VAE: A Spherical Vector Quantization Framework with ArcCosine Additive Margin

    cs.CV 2026-05 unverdicted novelty 5.0 of 10

    ArcVQ-VAE adds spherical angular-margin regularization consisting of ball-bounded norms and arc-cosine margin loss to improve codebook utilization in VQ-VAE for image tasks.

  4. Derivations of rational vertex operator algebras are inner

    math.QA 2026-06 unverdicted novelty 4.0 of 10

    Every derivation of a simple rational CFT-type vertex operator algebra is an inner derivation.

  5. Derivations of rational vertex operator algebras are inner

    math.QA 2026-06 unverdicted novelty 3.0 of 10

    Every derivation of a simple rational VOA of CFT type is an inner derivation.

Pith tools