Official implementation of CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification

Last update: Dec 29, 2022

Overview

CrossViT

This repository is the official implementation of CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification. ArXiv

If you use the codes and models from this repo, please cite our work. Thanks!

@inproceedings{
    chen2021crossvit,
    title={{CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification}},
    author={Chun-Fu (Richard) Chen and Quanfu Fan and Rameswar Panda},
    booktitle={International Conference on Computer Vision (ICCV)},
    year={2021}
}

Installation

To install requirements:

pip install -r requirements.txt

With conda:

conda create -n crossvit python=3.8
conda activate crossvit
conda install pytorch=1.7.1 torchvision  cudatoolkit=11.0 -c pytorch -c nvidia
pip install -r requirements.txt

Data preparation

Download and extract ImageNet train and val images from http://image-net.org/. The directory structure is the standard layout for the torchvision datasets.ImageFolder, and the training and validation data is expected to be in the train/ folder and val folder respectively:

/path/to/imagenet/
  train/
    class1/
      img1.jpeg
    class2/
      img2.jpeg
  val/
    class1/
      img3.jpeg
    class/2
      img4.jpeg

Pretrained models

We provide models trained on ImageNet1K. You can find models here. And you can load pretrained weights into models by add --pretrained flag.

Training

To train crossvit_9_dagger_224 on ImageNet on a single node with 8 gpus for 300 epochs run:

python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --model crossvit_9_dagger_224 --batch-size 256 --data-path /path/to/imagenet

Other model names can be found at models/crossvit.py.

Multinode training

Distributed training is available via Slurm and submitit:

To train a crossvit_9_dagger_224 model on ImageNet on 4 nodes with 8 gpus each for 300 epochs:

python run_with_submitit.py --nodes 4 --model crossvit_9_dagger_224 --data-path /path/to/imagenet --batch-size 128 --warmup-epochs 30

Or you can start process on each machine maunally. E.g. 2 nodes, each with 8 gpus.

Machine 0:

python -m torch.distributed.launch --nproc_per_node=8 --master_addr=MACHINE_0_IP --master_port=AVAILABLE_PORT --nnodes=2 --node_rank=0 main.py --model crossvit_9_dagger_224 --batch-size 256 --data-path /path/to/imagenet

Machine 1:

python -m torch.distributed.launch --nproc_per_node=8 --master_addr=MACHINE_0_IP --master_port=AVAILABLE_PORT --nnodes=2 --node_rank=1 main.py --model crossvit_9_dagger_224 --batch-size 256 --data-path /path/to/imagenet

Note that: some slurm configurations might need to be changed based on your cluster.

Evaluation

To evaluate a pretrained model on crossvit_9_dagger_224:

python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --model crossvit_9_dagger_224 --batch-size 128 --data-path /path/to/imagenet --eval --pretrained

Comments

Multilabel classification
@jjasghar @krook @chunfuchen thanks a lot for sharing the code , i have a problem statement which i want to train crossvit on please find the details below

input is of image shape 256x128 having label vector as gt for multilabel classification

can i use crossvit to train the model , what all modification has to be done with the code base ??

THnaks for the support
opened by abhigoku10 4
Honoring distributed flag + fixing reset_classifier
Honoring the args.distributed flag in calls to evaluate().

A couple of changes to make the reset_classifier() method work:

Initializing the embed_dim instance variable in VisionTransformer.

Reinitializing the classification head for all branches.
opened by abhrac 1
Parameter setting

Hello, thank you for your excellent work, I would like to know how you set the parameters on the CIFAR10 dataset, mainly the size of the patch，Looking forward to your reply

opened by happy20200 1

This repository builds a basic vision transformer from scratch so that one beginner can understand the theory of vision transformer.

vision-transformer-from-scratch This repository includes several kinds of vision transformers from scratch so that one beginner can understand the the

1 Dec 24, 2021

Code of U2Fusion: a unified unsupervised image fusion network for multiple image fusion tasks, including multi-modal, multi-exposure and multi-focus image fusion.

U2Fusion Code of U2Fusion: a unified unsupervised image fusion network for multiple image fusion tasks, including multi-modal (VIS-IR, medical), multi

129 Dec 11, 2022

This is an official implementation for "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" on Object Detection and Instance Segmentation.

Swin Transformer for Object Detection This repo contains the supported code and configuration files to reproduce object detection results of Swin Tran

1.4k Dec 30, 2022

Unofficial implementation of MUSIQ (Multi-Scale Image Quality Transformer)

MUSIQ: Multi-Scale Image Quality Transformer Unofficial pytorch implementation of the paper "MUSIQ: Multi-Scale Image Quality Transformer" (paper link

41 Jan 2, 2023

Official implementation of "Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets" (CVPR2021)

Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets This is the official implementation of "Towards Good Pract

52 Nov 22, 2022

Official implementation of paper "Query2Label: A Simple Transformer Way to Multi-Label Classification".

Introdunction This is the official implementation of the paper "Query2Label: A Simple Transformer Way to Multi-Label Classification". Abstract This pa

274 Dec 28, 2022

Implementation of STAM (Space Time Attention Model), a pure and simple attention model that reaches SOTA for video classification

STAM - Pytorch Implementation of STAM (Space Time Attention Model), yet another pure and simple SOTA attention model that bests all previous models in

109 Dec 28, 2022

Pytorch implementation of ICASSP 2022 paper Attention Probe: Vision Transformer Distillation in the Wild

Attention Probe: Vision Transformer Distillation in the Wild Jiahao Wang, Mingdeng Cao, Shuwei Shi, Baoyuan Wu, Yujiu Yang In ICASSP 2022 This code is

6 Sep 21, 2022

Implementation of a memory efficient multi-head attention as proposed in the paper, "Self-attention Does Not Need O(n²) Memory"

Memory Efficient Attention Pytorch Implementation of a memory efficient multi-head attention as proposed in the paper, Self-attention Does Not Need O(

180 Jan 5, 2023

Official implementation of CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification

Related tags

Overview

CrossViT

Installation

Data preparation

Pretrained models

Training

Multinode training

Evaluation

You might also like...

This repository builds a basic vision transformer from scratch so that one beginner can understand the theory of vision transformer.

Code of U2Fusion: a unified unsupervised image fusion network for multiple image fusion tasks, including multi-modal, multi-exposure and multi-focus image fusion.

This is an official implementation for "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" on Object Detection and Instance Segmentation.

Unofficial implementation of MUSIQ (Multi-Scale Image Quality Transformer)

Official implementation of "Towards Good Practices for Efficiently Annotating Large-Scale Image Classification Datasets" (CVPR2021)

Official implementation of paper "Query2Label: A Simple Transformer Way to Multi-Label Classification".

Implementation of STAM (Space Time Attention Model), a pure and simple attention model that reaches SOTA for video classification

Pytorch implementation of ICASSP 2022 paper Attention Probe: Vision Transformer Distillation in the Wild

Implementation of a memory efficient multi-head attention as proposed in the paper, "Self-attention Does Not Need O(n²) Memory"

Comments

Multilabel classification

Honoring distributed flag + fixing reset_classifier

Parameter setting

Releases(weights-0.1)

weights-0.1(Aug 14, 2021)

Owner

International Business Machines

A-SDF: Learning Disentangled Signed Distance Functions for Articulated Shape Representation (ICCV 2021)

ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators

Evolving neural network parameters in JAX.

Data and Code for ACL 2021 Paper "Inter-GPS: Interpretable Geometry Problem Solving with Formal Language and Symbolic Reasoning"

ELSED: Enhanced Line SEgment Drawing

ShinRL: A Library for Evaluating RL Algorithms from Theoretical and Practical Perspectives

Official PyTorch implementation of "Meta-Learning with Task-Adaptive Loss Function for Few-Shot Learning" (ICCV2021 Oral)

[CVPR 2021] Few-shot 3D Point Cloud Semantic Segmentation

A curated list of references for MLOps

CenterPoint 3D Object Detection and Tracking using center points in the bird-eye view.

Rational Activation Functions - Replacing Padé Activation Units

Learning to Disambiguate Strongly Interacting Hands via Probabilistic Per-Pixel Part Segmentation [3DV 2021 Oral]

Simple PyTorch implementations of Badnets on MNIST and CIFAR10.

A web porting for NVlabs' StyleGAN2, to facilitate exploring all kinds characteristic of StyleGAN networks

A Light CNN for Deep Face Representation with Noisy Labels

Code for paper "Vocabulary Learning via Optimal Transport for Neural Machine Translation"

Fair Recommendation in Two-Sided Platforms

Neurolab is a simple and powerful Neural Network Library for Python

Contrastive Language-Image Pretraining

Baseline inference Algorithm for the STOIC2021 challenge.