The implementation of "Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer"

Last update: Nov 29, 2022

Related tags

Overview

Shuffle Transformer

The implementation of "Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer"

Introduction

Very recently, window-based Transformers, which computed self-attention within non-overlapping local windows, demonstrated promising results on image classification, semantic segmentation, and object detection. However, less study has been devoted to the cross-window connection which is the key element to improve the representation ability. Shuffle Transformer revisit the spatial shuffle as an efficient way to build connections among windows, which is highly efficient and easy to implement by modifying two lines of code. Furthermore, the depth-wise convolution is introduced to complement the spatial shuffle for enhancing neighbor-window connections. The proposed architectures achieve excellent performance on a wide range of visual tasks including image-level classification, object detection, and semantic segmentation.

Requirements

PyTorch==1.7.1
torchvision==0.8.2
timm==0.3.2

The Apex is optional for faster training speed.

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

Other Requirements

pip install opencv-python==4.4.0.46 termcolor==1.1.0 yacs==0.1.8
pip install einops

Main Results

Results on ImageNet-1K

name	acc@1	#params	FLOPs	Throughputs(Images/s)	Weights
Shuffle-T	82.4	28M	4.6G	791	google drive
Shuffle-S	83.6	50M	8.9G	450	google drive
Shuffle-B	84.0	88M	15.6	279	google drive

Usage

For classification on ImageNet-1K, to train from scratch, run:

python -m torch.distributed.launch --nproc_per_node   main.py \ 
--cfg  --data-path  [--batch-size  --output ]

To evaluate, run:

python -m torch.distributed.launch --nproc_per_node  main.py --eval \
--cfg  --resume  --data-path

In progress

Semantic Segmentation
Instance Segmentation

Citing Shuffle Transformer

@article{huang2021shuffle,
 title={Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer},
 author={Huang, Zilong and Ben, Youcheng and Luo, Guozhong and Cheng, Pei and Yu, Gang and Fu, Bin},
 journal={arXiv preprint arXiv:2106.03650},
 year={2021}
}

Acknowledgement

Thanks to open-source implementation of Swin-Transformer.

Comments

关于shuffle操作代码的疑问

您好，请问在模型实现代码中为什么只需要换一下ws1和hh、ws2和ww的位置就能实现shuffle的操作呢，在括号里相乘(AxB)和(BxA)应该没有什么区别吧，期待您的解答！

就是下面的代码 q, k, v = rearrange(qkv, 'b (qkv h d) (ws1 hh) (ws2 ww) -> qkv (b hh ww) h (ws1 ws2) d', h=self.num_heads, qkv=3, ws1=self.ws, ws2=self.ws) q, k, v = rearrange(qkv, 'b (qkv h d) (hh ws1) (ww ws2) -> qkv (b hh ww) h (ws1 ws2) d', h=self.num_heads, qkv=3, ws1=self.ws, ws2=self.ws)

opened by endaoguansanlu 0
Any suggestions about shuffle-large design and training on larger dataset like ImageNet-22K?

I write a shuffle-large config following swin-large and training on ImageNet22K dataset using apex O1. But the training process is unstable and the loss quickly become NAN. Is there any suggestions about shuffle-large design and training on larger dataset like ImageNet-22K?

opened by jiandan42 0

get_sinusoid_encoding()

    if self.has_pos_embed:
        self.pos_embed = nn.Parameter(data=get_sinusoid_encoding(n_position=num_patches, d_hid=embed_dim), requires_grad=False)
        self.pos_drop = nn.Dropout(p=drop_rate)

Hello, is get_sinusoid_encoding used? Can I remove it if I don't use it? This function doesn't seem to be defined.

opened by Dpw506 1

🌳 A Python-inspired implementation of the Optimum-Path Forest classifier.

OPFython: A Python-Inspired Optimum-Path Forest Classifier Welcome to OPFython. Note that this implementation relies purely on the standard LibOPF. Th

30 Jan 4, 2023

Implementation of Geometric Vector Perceptron, a simple circuit for 3d rotation equivariance for learning over large biomolecules, in Pytorch. Idea proposed and accepted at ICLR 2021

Geometric Vector Perceptron Implementation of Geometric Vector Perceptron, a simple circuit with 3d rotation equivariance for learning over large biom

59 Nov 24, 2022

Official implementation of AAAI-21 paper "Label Confusion Learning to Enhance Text Classification Models"

Description: This is the official implementation of our AAAI-21 accepted paper Label Confusion Learning to Enhance Text Classification Models. The str

101 Nov 25, 2022

Official PyTorch implementation for paper Context Matters: Graph-based Self-supervised Representation Learning for Medical Images

Context Matters: Graph-based Self-supervised Representation Learning for Medical Images Official PyTorch implementation for paper Context Matters: Gra

49 Nov 23, 2022

PyTorch implementation of "Conformer: Convolution-augmented Transformer for Speech Recognition" (INTERSPEECH 2020)

PyTorch implementation of Conformer: Convolution-augmented Transformer for Speech Recognition. Transformer models are good at capturing content-based

565 Jan 4, 2023

An essential implementation of BYOL in PyTorch + PyTorch Lightning

Essential BYOL A simple and complete implementation of Bootstrap your own latent: A new approach to self-supervised Learning in PyTorch + PyTorch Ligh

48 Sep 27, 2022

The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf

NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021] Release Notes The offical PyTorch implementation of NeMo, p

76 Nov 23, 2022

A PyTorch re-implementation of the paper 'Exploring Simple Siamese Representation Learning'. Reproduced the 67.8% Top1 Acc on ImageNet.

Exploring simple siamese representation learning This is a PyTorch re-implementation of the SimSiam paper on ImageNet dataset. The results match that

72 Nov 9, 2022

PyTorch implementation of "A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement."

FullSubNet This Git repository for the official PyTorch implementation of "A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech E

357 Jan 4, 2023

The implementation of "Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer"

Related tags

Overview

Shuffle Transformer

Introduction

Requirements

Main Results

Results on ImageNet-1K

Usage

In progress

Citing Shuffle Transformer

Acknowledgement

You might also like...

🌳 A Python-inspired implementation of the Optimum-Path Forest classifier.

Implementation of Geometric Vector Perceptron, a simple circuit for 3d rotation equivariance for learning over large biomolecules, in Pytorch. Idea proposed and accepted at ICLR 2021

Official implementation of AAAI-21 paper "Label Confusion Learning to Enhance Text Classification Models"

Official PyTorch implementation for paper Context Matters: Graph-based Self-supervised Representation Learning for Medical Images

PyTorch implementation of "Conformer: Convolution-augmented Transformer for Speech Recognition" (INTERSPEECH 2020)

An essential implementation of BYOL in PyTorch + PyTorch Lightning

The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf

A PyTorch re-implementation of the paper 'Exploring Simple Siamese Representation Learning'. Reproduced the 67.8% Top1 Acc on ImageNet.

PyTorch implementation of "A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement."

Comments

关于shuffle操作代码的疑问

Any suggestions about shuffle-large design and training on larger dataset like ImageNet-22K?

get_sinusoid_encoding()

Owner

ALBERT-pytorch-implementation - ALBERT pytorch implementation

Numenta Platform for Intelligent Computing is an implementation of Hierarchical Temporal Memory (HTM), a theory of intelligence based strictly on the neuroscience of the neocortex.

PyTorch implementation of neural style transfer algorithm

PyTorch implementation of DeepDream algorithm

The project is an official implementation of our CVPR2019 paper "Deep High-Resolution Representation Learning for Human Pose Estimation"

Image-to-Image Translation with Conditional Adversarial Networks (Pix2pix) implementation in keras

Python implementation of cover trees, near-drop-in replacement for scipy.spatial.kdtree

Home repository for the Regularized Greedy Forest (RGF) library. It includes original implementation from the paper and multithreaded one written in C++, along with various language-specific wrappers.

Implementation of Restricted Boltzmann Machine (RBM) and its variants in Tensorflow

A fast Evolution Strategy implementation in Python