Pith. sign in

REVIEW 2 cited by

Early Convolutions Help Transformers See Better

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 2106.14881 v3 pith:XIUH7MZC submitted 2021-06-28 cs.CV

classification cs.CV
keywords stemconvolutionalmodelschoicedesignmodeloptimizationaccuracy
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Vision transformer (ViT) models exhibit substandard optimizability. In particular, they are sensitive to the choice of optimizer (AdamW vs. SGD), optimizer hyperparameters, and training schedule length. In comparison, modern convolutional neural networks are easier to optimize. Why is this the case? In this work, we conjecture that the issue lies with the patchify stem of ViT models, which is implemented by a stride-p p*p convolution (p=16 by default) applied to the input image. This large-kernel plus large-stride convolution runs counter to typical design choices of convolutional layers in neural networks. To test whether this atypical design choice causes an issue, we analyze the optimization behavior of ViT models with their original patchify stem versus a simple counterpart where we replace the ViT stem by a small number of stacked stride-two 3*3 convolutions. While the vast majority of computation in the two ViT designs is identical, we find that this small change in early visual processing results in markedly different training behavior in terms of the sensitivity to optimization settings as well as the final model accuracy. Using a convolutional stem in ViT dramatically increases optimization stability and also improves peak performance (by ~1-2% top-1 accuracy on ImageNet-1k), while maintaining flops and runtime. The improvement can be observed across the wide spectrum of model complexities (from 1G to 36G flops) and dataset scales (from ImageNet-1k to ImageNet-21k). These findings lead us to recommend using a standard, lightweight convolutional stem for ViT models in this regime as a more robust architectural choice compared to the original ViT model design.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 352 citations worldwide. Full citation record

  1. Self-Supervised Multisensory Pretraining for Contact-Rich Robot Reinforcement Learning

    cs.RO 2025-11 unverdicted novelty 6.0 of 10

    MSDP pre-trains a transformer encoder with masked multisensory autoencoding, then uses an asymmetric actor-critic bridge (cross-attention for critic, pooling for actor) to accelerate and robustify contact-rich RL acro...

  2. Faithful, Interpretable Chest X-ray Diagnosis with Anti-Aliased B-cos Networks

    cs.CV 2025-07 conditional novelty 4.0 of 10

    Combining B-cos networks with anti-aliasing pooling (FLC or BlurPool) reduces grid artifacts in chest X-ray explanation maps while keeping diagnostic accuracy close to baseline networks.

Pith tools