This repository contains an implementation of ConvMixer for the ICLR 2022 submission "Patches Are All You Need?".

Overview

Patches Are All You Need? 🤷

This repository contains an implementation of ConvMixer for the ICLR 2022 submission "Patches Are All You Need?".

Code overview

The most important code is in convmixer.py. We trained ConvMixers using the timm framework, which we copied from here.

Update: ConvMixer is now integrated into the timm framework itself. You can see the PR here.

Inside pytorch-image-models, we have made the following modifications. (Though one could look at the diff, we think it is convenient to summarize them here.)

  • Added ConvMixers
    • added timm/models/convmixer.py
    • modified timm/models/__init__.py
  • Added "OneCycle" LR Schedule
    • added timm/scheduler/onecycle_lr.py
    • modified timm/scheduler/scheduler.py
    • modified timm/scheduler/scheduler_factory.py
    • modified timm/scheduler/__init__.py
    • modified train.py (added two lines to support this LR schedule)

We are confident that the use of the OneCycle schedule here is not critical, and one could likely just as well train ConvMixers with the built-in cosine schedule.

Evaluation

We provide some model weights below:

Model Name Kernel Size Patch Size File Size
ConvMixer-1536/20 9 7 207MB
ConvMixer-768/32* 7 7 85MB
ConvMixer-1024/20 9 14 98MB

* Important: ConvMixer-768/32 here uses ReLU instead of GELU, so you would have to change convmixer.py accordingly (we will fix this later).

You can evaluate ConvMixer-1536/20 as follows:

python validate.py --model convmixer_1536_20 --b 64 --num-classes 1000 --checkpoint [/path/to/convmixer_1536_20_ks9_p7.pth.tar] [/path/to/ImageNet1k-val]

You should get a 81.37% accuracy.

Training

If you had a node with 10 GPUs, you could train a ConvMixer-1536/20 as follows (these are exactly the settings we used):

sh distributed_train.sh 10 [/path/to/ImageNet1k] 
    --train-split [your_train_dir] 
    --val-split [your_val_dir] 
    --model convmixer_1536_20 
    -b 64 
    -j 10 
    --opt adamw 
    --epochs 150 
    --sched onecycle 
    --amp 
    --input-size 3 224 224
    --lr 0.01 
    --aa rand-m9-mstd0.5-inc1 
    --cutmix 0.5 
    --mixup 0.5 
    --reprob 0.25 
    --remode pixel 
    --num-classes 1000 
    --warmup-epochs 0 
    --opt-eps=1e-3 
    --clip-grad 1.0

We also included a ConvMixer-768/32 in timm/models/convmixer.py (though it is simple to add more ConvMixers). We trained that one with the above settings but with 300 epochs instead of 150 epochs.

In the near future, we will upload weights.

The tweetable version of ConvMixer, which requires from torch.nn import *:

def ConvMixr(h,d,k,p,n):
 S,C,A=Sequential,Conv2d,lambda x:S(x,GELU(),BatchNorm2d(h))
 R=type('',(S,),{'forward':lambda s,x:s[0](x)+x})
 return S(A(C(3,h,p,p)),*[S(R(A(C(h,h,k,groups=h,padding=k//2))),A(C(h,h,1))) for i in range(d)],AdaptiveAvgPool2d((1,1)),Flatten(),Linear(h,n))
Comments
  • Cifar10 baseline doesn't reach 95%

    Cifar10 baseline doesn't reach 95%

    Hello, I tried convmixer256 on Cifar-10 with the same timm options specified for ImageNet (except the num_classes) and it doesn't go beyond 90% accuracy. Could you please specify the options used for Cifar-10 experiment ?

    opened by K-H-Ismail 13
  • What's new about this model?

    What's new about this model?

    Why “patches” are all you need? Patch embedding is Conv7x7 stem, The body is simply repeated Conv9x9 + Conv1x1, (Not challenging your work, it's indeed very interesting), but just kindly wondering what's new about this model?

    opened by vztu 5
  • Training scheme modifications for small GPUs

    Training scheme modifications for small GPUs

    Hi authors. Your paper has demonstrated a quite intriguing observation. I wish you luck with your submission. Thanks for sharing the code of the submission. When running the code, I got an issue regarding OOM when using the default batch size of 64. In the end I can only run with 8 samples per batch per GPU as my GPUs have only 11GB. I would like to know if you have tried smaller GPUs and achieved the same results. So far, besides learning rate modified according to the linear rule, I haven't made any change yet. If you tried training using smaller GPUs before, could you please share your experience? Thank you very much!

    opened by justanhduc 4
  • Experiments with full convolutional layers instead of patch embedding?

    Experiments with full convolutional layers instead of patch embedding?

    Have the author tried to replace the patch embedding with the just convolution?That is, using 1 stride instead of p?

    With this setting, this is a standard convolution network like MobileNet. I wonder what would be the performance?Is the performance gain of Convmix due to the patch embedding or the depthwise conv layers?

    Very interested in this work, thanks.

    opened by forjiuzhou 2
  • Training time

    Training time

    Hi, first of all thanks for a very interesting paper.

    I would like to know how long did it take you to train the models? I'm trying to train ConvMixer-768/32 using 2xV100 and one epoch is ~3 hours, so I would estimate that full training would take ~= 2 * 3 * 300 ~= 1800 GPU hours, which is insane. Even if you trained with 10 GPUs it would take ~1 week for one experiment to finish. Are my calculations correct?

    opened by bonlime 1
  • padding=same?

    padding=same?

    https://github.com/tmp-iclr/convmixer/blob/1cefd860a1a6a85369887d1a633425cedc2afd0a/convmixer.py#L18 There is an error:TypeError: conv2d(): argument 'padding' (position 5) must be tuple of ints, not str.

    opened by linhaoqi027 1
  • Add Docker environment & web demo

    Add Docker environment & web demo

    Hey @ashertrockman, @tmp-iclr ! wave

    This pull request makes it possible to run your model inside a Docker environment, which makes it easier for other people to run it. We're using an open source tool called Cog to make this process easier.

    This also means we can make a web page where other people can try out your model! View it here: https://replicate.com/locuslab/convmixer and have a look at some Image classification examples we already uploaded.

    By clicking "Claim this model" You'll be able to edit the everything, and we'll feature it on our website and tweet about it too.

    In case you're wondering who I am, I'm from Replicate, where we're trying to make machine learning reproducible. We got frustrated that we couldn't run all the really interesting ML work being done. So, we're going round implementing models we like. blush

    opened by ariel415el 0
  • Add Docker environment & web demo

    Add Docker environment & web demo

    Hey @ashertrockman, @tmp-iclr ! 👋

    This pull request makes it possible to run your model inside a Docker environment, which makes it easier for other people to run it. We're using an open source tool called Cog to make this process easier.

    This also means we can make a web page where other people can try out your model! View it here: https://replicate.com/locuslab/convmixer and have a look at some Image classification examples we already uploaded.

    By clicking "Claim this model" You'll be able to edit the everything, and we'll feature it on our website and tweet about it too.

    In case you're wondering who I am, I'm from Replicate, where we're trying to make machine learning reproducible. We got frustrated that we couldn't run all the really interesting ML work being done. So, we're going round implementing models we like. 😊

    opened by ariel415el 0
  • Fix notebooks

    Fix notebooks

    Hi.

    Fixed errors in pytorch-image-models/notebooks/{EffResNetComparison,GeneralizationToImageNetV2}.ipynb notebooks:

    • added missed pynvml installation;
    • resolved missed imports;
    • resolved errors due to outdated calls of timm library.

    Tested in colab env: "Run all" without any errors.

    opened by amrzv 0
  • CIFAR-10 training settings

    CIFAR-10 training settings

    First of all, thank you for the interesting work. I was experimenting the one with patch size 1 and kernel size 9 with CIFAR-10 with the following training settings:

    --model tiny_convmixer
     -b 64 -j 8 
    --opt adamw 
    --epochs 200 
    --sched onecycle 
    --amp 
    --input-size 3 32 32 
    --lr 0.01 
    --aa rand-m9-mstd0.5-inc1 
    --cutmix 0.5 
    --mixup 0.5 
    --reprob 0.25 
    --remode pixel 
    --num-classes 10
    --warmup-epochs 0
    --opt-eps 1e-3
    --clip-grad 1.0
    --scale 0.75 1.0
    --weight-decay 0.01
    --mean 0.4914 0.4822 0.4465
    --std 0.2471 0.2435 0.2616
    

    I could get only 95.89%. I am supposed to get 96.03% according to Table 4 in the paper. Can you please let me know any setting I missed? Thank you again.

    opened by fugokidi 0
  • Segmentation ConvMixer architecture ?

    Segmentation ConvMixer architecture ?

    I was trying to figure what a Segmentation ConvMixer would look like, and came up with that (residual connection inspired by MultiResUNet). Does it make sense to you ?

    image

    opened by divideconcept 0
  • Request more experiment results to compare to other architecture.

    Request more experiment results to compare to other architecture.

    Hi! This work is pretty interesting, but I think there should are more results like in "Demystifying Local Vision Transformer: Sparse Connectivity, Weight Sharing, and Dynamic Weight" as they replace local self-attention with depth-wise convolution in Swin Transformer. Since you conduct an advanced one with a more simple architecture compared to SwinTransformer, so I wonder if ConvMixer can get similar performance on object detection and semantic segmentation.

    opened by LuoXin-s 1
Releases(timm-v1.0)
Owner
ICLR 2022 Author
Patches Are All You Need? 🤷
ICLR 2022 Author
AI4Good project for detecting waste in the environment

Detect waste AI4Good project for detecting waste in environment. www.detectwaste.ml. Our latest results were published in Waste Management journal in

108 Dec 25, 2022
Code for "Hierarchical Skills for Efficient Exploration" HSD-3 Algorithm and Baselines

Hierarchical Skills for Efficient Exploration This is the source code release for the paper Hierarchical Skills for Efficient Exploration. It contains

Facebook Research 38 Dec 06, 2022
Sync2Gen Code for ICCV 2021 paper: Scene Synthesis via Uncertainty-Driven Attribute Synchronization

Sync2Gen Code for ICCV 2021 paper: Scene Synthesis via Uncertainty-Driven Attribute Synchronization 0. Environment Environment: python 3.6 and cuda 10

Haitao Yang 62 Dec 30, 2022
PyTorch implementation of ICLR 2022 paper PiCO: Contrastive Label Disambiguation for Partial Label Learning

PiCO: Contrastive Label Disambiguation for Partial Label Learning This is a PyTorch implementation of ICLR 2022 Oral paper PiCO; also see our Project

王皓波 147 Jan 07, 2023
PyTorch version repo for CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes

Study-CSRNet-pytorch This is the PyTorch version repo for CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes

0 Mar 01, 2022
Brain tumor detection using Convolution-Neural Network (CNN)

Detect and Classify Brain Tumor using CNN. A system performing detection and classification by using Deep Learning Algorithms using Convolution-Neural Network (CNN).

assia 1 Feb 07, 2022
End-to-End Dense Video Captioning with Parallel Decoding (ICCV 2021)

PDVC Official implementation for End-to-End Dense Video Captioning with Parallel Decoding (ICCV 2021) [paper] [valse论文速递(Chinese)] This repo supports:

Teng Wang 118 Dec 16, 2022
A fast python implementation of Ray Tracing in One Weekend using python and Taichi

ray-tracing-one-weekend-taichi A fast python implementation of Ray Tracing in One Weekend using python and Taichi. Taichi is a simple "Domain specific

157 Dec 26, 2022
Custom implementation of Corrleation Module

Pytorch Correlation module this is a custom C++/Cuda implementation of Correlation module, used e.g. in FlowNetC This tutorial was used as a basis for

Clément Pinard 361 Dec 12, 2022
Spectralformer: Rethinking hyperspectral image classification with transformers

Spectralformer: Rethinking hyperspectral image classification with transformers Danfeng Hong, Zhu Han, Jing Yao, Lianru Gao, Bing Zhang, Antonio Plaza

Danfeng Hong 102 Dec 29, 2022
This repository contains datasets and baselines for benchmarking Chinese text recognition.

Benchmarking-Chinese-Text-Recognition This repository contains datasets and baselines for benchmarking Chinese text recognition. Please see the corres

FudanVI Lab 254 Dec 30, 2022
Diffusion Probabilistic Models for 3D Point Cloud Generation (CVPR 2021)

Diffusion Probabilistic Models for 3D Point Cloud Generation [Paper] [Code] The official code repository for our CVPR 2021 paper "Diffusion Probabilis

Shitong Luo 323 Jan 05, 2023
NeRF Meta-Learning with PyTorch

NeRF Meta Learning With PyTorch nerf-meta is a PyTorch re-implementation of NeRF experiments from the paper "Learned Initializations for Optimizing Co

Sanowar Raihan 78 Dec 18, 2022
Convert human motion from video to .bvh

video_to_bvh Convert human motion from video to .bvh with Google Colab Usage 1. Open video_to_bvh.ipynb in Google Colab Go to https://colab.research.g

Dene 306 Dec 10, 2022
RoFormer_pytorch

PyTorch RoFormer 原版Tensorflow权重(https://github.com/ZhuiyiTechnology/roformer) chinese_roformer_L-12_H-768_A-12.zip (提取码:xy9x) 已经转化为PyTorch权重 chinese_r

yujun 283 Dec 12, 2022
Self-supervised Multi-modal Hybrid Fusion Network for Brain Tumor Segmentation

JBHI-Pytorch This repository contains a reference implementation of the algorithms described in our paper "Self-supervised Multi-modal Hybrid Fusion N

FeiyiFANG 5 Dec 13, 2021
Large-Scale Unsupervised Object Discovery

Large-Scale Unsupervised Object Discovery Huy V. Vo, Elena Sizikova, Cordelia Schmid, Patrick Pérez, Jean Ponce [PDF] We propose a novel ranking-based

17 Sep 19, 2022
Adversarial Self-Defense for Cycle-Consistent GANs

Adversarial Self-Defense for Cycle-Consistent GANs This is the official implementation of the CycleGAN robust to self-adversarial attacks used in pape

Dina Bashkirova 10 Oct 10, 2022
ShapeGlot: Learning Language for Shape Differentiation

ShapeGlot: Learning Language for Shape Differentiation Created by Panos Achlioptas, Judy Fan, Robert X.D. Hawkins, Noah D. Goodman, Leonidas J. Guibas

Panos 32 Dec 23, 2022