CAIRI Supervised, Semi- and Self-Supervised Visual Representation Learning Toolbox and Benchmark
📘Documentation | 🛠️Installation | 🚀Model Zoo | 👀Awesome Mixup | 🔍Awesome MIM | 🆕News
The main branch works with PyTorch 1.8 (required by some self-supervised methods) or higher (we recommend PyTorch 1.12). You can still use PyTorch 1.6 for supervised classification methods.
OpenMixup
is an open-source toolbox for supervised, self-, and semi-supervised visual representation learning with mixup based on PyTorch, especially for mixup-related methods. Recently, OpenMixup
is on updating to adopt new features and code structures of OpenMMLab 2.0 (#42).
Modular Design. OpenMixup follows a similar code architecture of OpenMMLab projects, which decompose the framework into various components, and users can easily build a customized model by combining different modules. OpenMixup is also transplantable to OpenMMLab projects (e.g., MMPreTrain).
All in One. OpenMixup provides popular backbones, mixup methods, semi-supervised, and self-supervised algorithms. Users can perform image classification (CNN & Transformer) and self-supervised pre-training (contrastive and autoregressive) under the same framework.
Standard Benchmarks. OpenMixup supports standard benchmarks of image classification, mixup classification, self-supervised evaluation, and provides smooth evaluation on downstream tasks with open-source projects (e.g., object detection and segmentation on Detectron2 and MMSegmentation).
State-of-the-art Methods. Openmixup provides awesome lists of popular mixup and self-supervised methods. OpenMixup is updating to support more state-of-the-art image classification and self-supervised methods.
[2023-12-23] OpenMixup
v0.2.9 is released, updating more features in mixup augmentations, self-supervised learning, and optimizers.
OpenMixup is compatible with Python 3.6/3.7/3.8/3.9 and PyTorch >= 1.6. Here are quick installation steps for development:
conda create -n openmixup python=3.8 pytorch=1.12 cudatoolkit=11.3 torchvision -c pytorch -y
conda activate openmixup
pip install openmim
mim install mmcv-full
git clone https://github.com/Westlake-AI/openmixup.git
cd openmixup
python setup.py develop
Please refer to install.md for more detailed installation and dataset preparation.
OpenMixup supports Linux and macOS. It enables easy implementation and extensions of mixup data augmentation methods in existing supervised, self-, and semi-supervised visual recognition models. Please see get_started.md for the basic usage of OpenMixup.
Here, we provide scripts for starting a quick end-to-end training with multiple GPUs
and the specified CONFIG_FILE
.
bash tools/dist_train.sh ${CONFIG_FILE} ${GPUS} [optional arguments]
For example, you can run the script below to train a ResNet-50 classifier on ImageNet with 4 GPUs:
CUDA_VISIBLE_DEVICES=0,1,2,3 PORT=29500 bash tools/dist_train.sh configs/classification/imagenet/resnet/resnet50_4xb64_cos_ep100.py 4
After training, you can test the trained models with the corresponding evaluation script:
bash tools/dist_test.sh ${CONFIG_FILE} ${GPUS} ${PATH_TO_MODEL} [optional arguments]
Please see Tutorials for more developing examples and tech details:
Downetream Tasks for Self-supervised Learning
Useful Tools
Please run experiments or find results on each config page. Refer to Mixup Benchmarks for benchmarking results of mixup methods. View Model Zoos Sup and Model Zoos SSL for a comprehensive collection of mainstream backbones and self-supervised algorithms. We also provide the paper lists of Awesome Mixups and Awesome MIM for your reference. Please view config files and links to models at the following config pages. Checkpoints and training logs are on updating!
Backbone architectures for supervised image classification on ImageNet.
Mixup methods for supervised image classification.
Self-supervised algorithms for visual representation learning.
Please refer to changelog.md for more details and release history.
This project is released under the Apache 2.0 license. See LICENSE
for more information.
If you find this project useful in your research, please consider star OpenMixup
or cite our tech report:
@article{li2022openmixup,
title = {OpenMixup: A Comprehensive Mixup Benchmark for Visual Classification},
author = {Siyuan Li and Zedong Wang and Zicheng Liu and Di Wu and Cheng Tan and Stan Z. Li},
journal = {ArXiv},
year = {2022},
volume = {abs/2209.04851}
}
For help, new features, or reporting bugs associated with OpenMixup, please open a GitHub issue and pull request with the tag "help wanted" or "enhancement". For now, the direct contributors include: Siyuan Li (@Lupin1998), Zedong Wang (@Jacky1128), and Zicheng Liu (@pone7). We thank all public contributors and contributors from MMPreTrain (MMSelfSup and MMClassification)!
This repo is currently maintained by: