PeCo: Perceptual Codebook for BERT Pre-training of Vision Transformers

Abstract

This paper explores a better prediction target for BERT pre-training of vision transformers. We observe that current prediction targets disagree with human perception judgment.This contradiction motivates us to learn a perceptual prediction target. We argue that perceptually similar images should stay close to each other in the prediction target space. We surprisingly find one simple yet effective idea: enforcing perceptual similarity during the dVAE training. Moreover, we adopt a self-supervised transformer model for deep feature extraction and show that it works well for calculating perceptual similarity.We demonstrate that such learned visual tokens indeed exhibit better semantic meanings, and help pre-training achieve superior transfer performance in various downstream tasks. For example, we achieve $\text{84.5\%}$ Top-1 accuracy on ImageNet-1K with ViT-B backbone, outperforming the competitive method BEiT by $\text{+1.3\%}$ under the same pre-training epochs. Our approach also gets significant improvement on object detection and segmentation on COCO and semantic segmentation on ADE20K. Equipped with a larger backbone ViT-H, we achieve the state-of-the-art ImageNet accuracy (\textbf{88.3\%}) among methods using only ImageNet-1K data.

arXiv commentTo appear at AAAI 2023

References in corpus (24)

• Adam: A Method for Stochastic Optimization
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Decoupled Weight Decay Regularization
• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• Bootstrap your own latent: A new approach to self-supervised Learning
• Language Models are Few-Shot Learners
• Neural Discrete Representation Learning
• Improved Baselines with Momentum Contrastive Learning
• Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• Pixel Recurrent Neural Networks
• Zero-Shot Text-to-Image Generation
• Unified Language Model Pre-training for Natural Language Understanding and Generation
• BEiT: BERT Pre-Training of Image Transformers
• MMDetection: Open MMLab Detection Toolbox and Benchmark
• Learning Representations by Maximizing Mutual Information Across Views
• Big Self-Supervised Models are Strong Semi-Supervised Learners
• Perceptual Adversarial Networks for Image-to-Image Transformation
• Stand-Alone Self-Attention in Vision Models
• Masked Autoencoders Are Scalable Vision Learners
• CSWin Transformer: A General Vision Transformer Backbone with Cross-Shaped Windows
• Super-Resolution with Deep Convolutional Sufficient Statistics
• Discrete Variational Autoencoders

Cited by in corpus (3)

• Contrastive Masked Autoencoders are Stronger Vision Learners
• CMID: A Unified Self-Supervised Learning Framework for Remote Sensing Image Understanding
• Towards Label-efficient Automatic Diagnosis and Analysis: A Comprehensive Survey of Advanced Deep Learning-based Weakly-supervised, Semi-supervised and Self-supervised Techniques in Histopathological Image Analysis