PSAQ-ViT V2: Toward Accurate and General Data-Free Quantization for Vision Transformers (2022-09-13T00:00:00.000000Z)

TL;DR

A more accurate and general data-free quantization framework for ViTs, built on top of PSAQ-ViT, and introduces an adaptive teacher–student strategy, which facilitates the constant cyclic evolution of the generated samples and the quantized model (student) in a competitive and interactive fashion under the supervision of the full-precision (FP) model (teacher), thus significantly improving the accuracy of the quantized model.

Abstract

Data-free quantization can potentially address data privacy and security concerns in model compression and thus has been widely investigated. Recently, patch similarity aware data-free quantization for vision transformers (PSAQ-ViT) designs a relative value metric, patch similarity, to generate data from pretrained vision transformers (ViTs), achieving the first attempt at data-free quantization for ViTs. In this article, we propose PSAQ-ViT V2, a more accurate and general data-free quantization framework for ViTs, built on top of PSAQ-ViT. More specifically, following the patch similarity metric in PSAQ-ViT, we introduce an adaptive teacher–student strategy, which facilitates the constant cyclic evolution of the generated samples and the quantized model (student) in a competitive and interactive fashion under the supervision of the full-precision (FP) model (teacher), thus significantly improving the accuracy of the quantized model. Moreover, without the auxiliary category guidance, we employ the task- and model-independent prior information, making the general-purpose scheme compatible with a broad range of vision tasks and models. Extensive experiments are conducted on various models on image classification, object detection, and semantic segmentation tasks, and PSAQ-ViT V2, with the naive quantization strategy and without access to real-world data, consistently achieves competitive results, showing potential as a powerful baseline on data-free quantization for ViTs. For instance, with Swin-S as the (backbone) model, 8-bit quantization reaches 82.13 top-1 accuracy on ImageNet, 50.9 box AP and 44.1 mask AP on COCO, and 47.2 mean Intersection over Union (mIoU) on ADE20K. We hope that accurate and general PSAQ-ViT V2 can serve as a potential and practice solution in real-world applications involving sensitive data. Code is released and merged at: https://github.com/zkkli/PSAQ-ViT.

References70 items

Dual-discriminator adversarial framework for data-free quantization

TinyViT: Fast Pretraining Distillation for Small Vision Transformers

An Effective Video Transformer With Synchronized Spatiotemporal and Spatial Self-Attention for Action Recognition

I-ViT: Integer-only Quantization for Efficient Vision Transformer Inference

Patch Similarity Aware Data-Free Quantization for Vision Transformers

SQuant: On-the-Fly Data-Free Quantization via Diagonal Hessian Approximation

A Vision Transformer Model for Convolution-Free Multilabel Classification of Satellite Imagery in Deforestation Monitoring

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

PTQ4ViT: Post-training Quantization for Vision Transformers with Twin Uniform Quantization

IntraQ: Learning Synthetic Images with Intra-Class Heterogeneity for Zero-Shot Network Quantization

Qimera: Data-free Quantization with Synthetic Boundary Supporting Samples

MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer

Rethinking and Improving Relative Position Encoding for Vision Transformer

Learning Efficient Vision Transformers via Fine-Grained Manifold Distillation

Post-Training Quantization for Vision Transformer

Patch Slimming for Efficient Vision Transformers

Annealing Knowledge Distillation

General Bitwidth Assignment for Efficient Deep Convolutional Neural Network Quantization

Transformer Tracking

ViViT: A Video Vision Transformer

A Survey of Quantization Methods for Efficient Neural Network Inference

DeepViT: Towards Deeper Vision Transformer

Diversifying Sample Generation for Accurate Data-Free Quantization

Transformer in Transformer

BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction

Video Transformer Network

Transformers in Vision: A Survey

A Survey on Vision Transformer

Training data-efficient image transformers & distillation through attention

Pre-Trained Image Processing Transformer

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Deformable DETR: Deformable Transformers for End-to-End Object Detection

One Weight Bitwidth to Rule Them All

Exploiting Retraining-Based Mixed-Precision Quantization for Low-Cost DNN Accelerator Design

Unsupervised Network Quantization via Fixed-Point Factorization

EasyQuant: Post-training Quantization via Scale Optimization

End-to-End Object Detection with Transformers

Bayesian Bits: Unifying Quantization and Pruning

Up or Down? Adaptive Rounding for Post-Training Quantization

Generative Low-bitwidth Data Free Quantization

Gradient-Based Deep Quantization of Neural Networks through Sinusoidal Adaptive Regularization

A Simple Framework for Contrastive Learning of Visual Representations

ZeroQ: A Novel Zero Shot Quantization Framework

Dreaming to Distill: Data-Free Knowledge Transfer via DeepInversion

Additive Powers-of-Two Quantization: An Efficient Non-uniform Discretization for Neural Networks

MMDetection: Open MMLab Detection Toolbox and Benchmark

Fully Quantized Network for Object Detection

Learned Step Size Quantization

Low-bit Quantization of Neural Networks for Efficient Inference

Unified Perceptual Parsing for Scene Understanding

LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks

Quantizing deep convolutional networks for efficient inference: A whitepaper

PACT: Parameterized Clipping Activation for Quantized Neural Networks

Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

Cascade R-CNN: Delving Into High Quality Object Detection

Semantic Understanding of Scenes Through the ADE20K Dataset

XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks

Adam: A Method for Stochastic Optimization

Microsoft COCO: Common Objects in Context

Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation

ImageNet: A large-scale hierarchical image database

FQ-ViT: Fully Quantized Vision Transformer without Retraining

Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

PTQ4ViT: Post-Training Quantization Framework for Vision Transformers

Efficient Vision Transformers via Fine-Grained Manifold Distillation

MMSegmentation: Openmmlab Semantic

received the B.Sc. degree from Xidian University, Shaanxi, China

Pytorch image models.

Binarized Neural Networks

The prior information used in this scheme is task-and model-independent, just in contrast to the category prior