MindCV is an open-source toolbox for computer vision research and development based on MindSpore. It collects a series of classic and SoTA vision models, such as ResNet and SwinTransformer, along with their pre-trained weights and training strategies. SoTA methods such as auto augmentation are also provided for performance improvement. With the decoupled module design, it is easy to apply or adapt MindCV to your own CV tasks.
The following is the corresponding mindcv versions and supported mindspore versions.
| mindcv | mindspore |
|---|---|
| main | master |
| v0.4.0 | 2.3.0 |
| 0.3.0 | 2.2.10 |
| 0.2 | 2.0 |
| 0.1 | 1.8 |
-
Easy-to-Use. MindCV decomposes the vision framework into various configurable components. It is easy to customize your data pipeline, models, and learning pipeline with MindCV:
>>> import mindcv # create a dataset >>> dataset = mindcv.create_dataset('cifar10', download=True) # create a model >>> network = mindcv.create_model('resnet50', pretrained=True)
Users can customize and launch their transfer learning or training task in one command line.
# transfer learning in one command line python train.py --model=swin_tiny --pretrained --opt=adamw --lr=0.001 --data_dir=/path/to/data -
State-of-The-Art. MindCV provides various CNN-based and Transformer-based vision models including SwinTransformer. Their pretrained weights and performance reports are provided to help users select and reuse the right model:
-
Flexibility and efficiency. MindCV is built on MindSpore which is an efficient DL framework that can be run on different hardware platforms (GPU/CPU/Ascend). It supports both graph mode for high efficiency and pynative mode for flexibility.
The performance of the models trained with MindCV is summarized in here, where the training recipes and weights are both available.
Model introduction and training details can be viewed in each sub-folder under configs.
See Installation for details.
To get started with MindCV, please see the Quick Start, which will give you a quick tour on each key component and the train/validate/predict pipelines.
Below are a few code snippets for your taste.
>>> import mindcv
# List and find a pretrained vision model
>>> mindcv.list_models("swin*", pretrained=True)
['swin_tiny']
# Create the model object
>>> network = mindcv.create_model('swin_tiny', pretrained=True)# Validate its accuracy
python validate.py --model=swin_tiny --pretrained --dataset=imagenet --val_split=validation
# {'Top_1_Accuracy': 0.80824, 'Top_5_Accuracy': 0.94802, 'loss': 1.7331367141008378}Image classification demo
Right click on the image below and save as dog.jpg.
Classify the downloaded image with a pretrained SoTA model:
python infer.py --model=swin_tiny --image_path='./dog.jpg'
# {'Labrador retriever': 0.5700152, 'golden retriever': 0.034551315, 'kelpie': 0.010108651, 'Chesapeake Bay retriever': 0.008229004, 'Walker hound, Walker foxhound': 0.007791956}The top-1 prediction result is labrador retriever, which is the breed of this cut dog.
It is easy to train your model on a standard or customized dataset using train.py, where the training strategy (e.g., augmentation, LR scheduling) can be configured with external arguments or a yaml config file.
-
Standalone Training
# standalone training python train.py --model=resnet50 --dataset=cifar10 --dataset_downloadAbove is an example for training ResNet50 on CIFAR10 dataset on a CPU/GPU/Ascend device
-
Distributed Training
For large datasets like ImageNet, it is necessary to do training in distributed mode on multiple devices. This can be achieved with
mpirunand parallel features supported by MindSpore.# distributed training # assume you have 4 GPUs/NPUs mpirun -n 4 python train.py --distribute \ --model=densenet121 --dataset=imagenet --data_dir=/path/to/imagenet
Notes: If the script is executed by the root user, the
--allow-run-as-rootparameter must be added tompirun.Detailed parameter definitions can be seen in
config.pyand checked by running `python train.py --help'.To resume training, please set the
--ckpt_pathand--ckpt_save_dirarguments. The optimizer state including the learning rate of the last stopped epoch will also be recovered. -
Config and Training Strategy
You can configure your model and other components either by specifying external parameters or by writing a yaml config file. Here is an example of training using a preset yaml file.
mpirun --allow-run-as-root -n 4 python train.py -c configs/squeezenet/squeezenet_1.0_gpu.yaml
Pre-defined Training Strategies: We provide more than 20 training recipes that achieve SoTA results on ImageNet currently. Please look into the
configsfolder for details. Please feel free to adapt these training strategies to your own model for performance improvement, which can be easily done by modifying the yaml file. -
Train on ModelArts/OpenI Platform
To run training on the ModelArts or OpenI cloud platform:
1. Create a new training task on the cloud platform. 2. Add run parameter `config` and specify the path to the yaml config file on the website UI interface. 3. Add run parameter `enable_modelarts` and set True on the website UI interface. 4. Fill in other blanks on the website and launch the training task.
Graph Mode and PyNative Mode:
By default, the training pipeline train.py is run in graph mode on MindSpore, which is optimized for efficiency and parallel computing with a compiled static graph.
In contrast, pynative mode is optimized for flexibility and easy debugging. You may alter the parameter --mode to switch to pure pynative mode for debugging purpose.
Mixed Mode:
PyNative mode with mindspore.jit is a mixed mode for comprising flexibility and efficiency in MindSpore. To apply pynative mode with mindspore.jit for training, please run train_with_func.py, e.g.,
python train_with_func.py --model=resnet50 --dataset=cifar10 --dataset_download --epoch_size=10Note: this is an experimental function under improvement. It is not stable on MindSpore 1.8.1 or earlier versions.
To evaluate the model performance, please run validate.py
# validate a trained checkpoint
python validate.py --model=resnet50 --dataset=imagenet --data_dir=/path/to/data --ckpt_path=/path/to/model.ckptValidation while Training
You can also track the validation accuracy during training by enabling the --val_while_train option.
python train.py --model=resnet50 --dataset=cifar10 \
--val_while_train --val_split=test --val_interval=1The training loss and validation accuracy for each epoch will be saved in {ckpt_save_dir}/results.log.
More examples about training and validation can be seen in examples.
We provide the following jupyter notebook tutorials to help users learn to use MindCV.
- Learn about configs
- Inference with a pretrained model
- Finetune a pretrained model on custom datasets
- Customize your model //coming soon
- Optimizing performance for vision transformer //coming soon
- Deployment demo
Currently, MindCV supports the model families listed below. More models with pre-trained weights are under development and will be released soon.
Supported models
- Big Transfer ResNetV2 (BiT) - https://arxiv.org/abs/1912.11370
- ConvNeXt - https://arxiv.org/abs/2201.03545
- ConViT (Soft Convolutional Inductive Biases Vision Transformers)- https://arxiv.org/abs/2103.10697
- DenseNet - https://arxiv.org/abs/1608.06993
- DPN (Dual-Path Network) - https://arxiv.org/abs/1707.01629
- EfficientNet (MBConvNet Family) https://arxiv.org/abs/1905.11946
- EfficientNet V2 - https://arxiv.org/abs/2104.00298
- GhostNet - https://arxiv.org/abs/1911.11907
- GoogLeNet - https://arxiv.org/abs/1409.4842
- Inception-V3 - https://arxiv.org/abs/1512.00567
- Inception-ResNet-V2 and Inception-V4 - https://arxiv.org/abs/1602.07261
- MNASNet - https://arxiv.org/abs/1807.11626
- MobileNet-V1 - https://arxiv.org/abs/1704.04861
- MobileNet-V2 - https://arxiv.org/abs/1801.04381
- MobileNet-V3 (MBConvNet w/ Efficient Head) - https://arxiv.org/abs/1905.02244
- NASNet - https://arxiv.org/abs/1707.07012
- PNasNet - https://arxiv.org/abs/1712.00559
- PVT (Pyramid Vision Transformer) - https://arxiv.org/abs/2102.12122
- PoolFormer models - https://github.com/sail-sg/poolformer
- RegNet - https://arxiv.org/abs/2003.13678
- RepMLP https://arxiv.org/abs/2105.01883
- RepVGG - https://arxiv.org/abs/2101.03697
- ResNet (v1b/v1.5) - https://arxiv.org/abs/1512.03385
- ResNeXt - https://arxiv.org/abs/1611.05431
- Res2Net - https://arxiv.org/abs/1904.01169
- ReXNet - https://arxiv.org/abs/2007.00992
- ShuffleNet v1 - https://arxiv.org/abs/1707.01083
- ShuffleNet v2 - https://arxiv.org/abs/1807.11164
- SKNet - https://arxiv.org/abs/1903.06586
- SqueezeNet - https://arxiv.org/abs/1602.07360
- Swin Transformer - https://arxiv.org/abs/2103.14030
- VGG - https://arxiv.org/abs/1409.1556
- Visformer - https://arxiv.org/abs/2104.12533
- Vision Transformer (ViT) - https://arxiv.org/abs/2010.11929
- Xception - https://arxiv.org/abs/1610.02357
Please see configs for the details about model performance and pretrained weights.
Supported algorithms
- Augmentation
- AutoAugment
- RandAugment
- Repeated Augmentation
- RandErasing (Cutout)
- CutMix
- MixUp
- RandomResizeCrop
- Color Jitter, Flip, etc
- Optimizer
- Adam
- AdamW
- Lion
- Adan (experimental)
- AdaGrad
- LAMB
- Momentum
- RMSProp
- SGD
- NAdam
- LR Scheduler
- Warmup Cosine Decay
- Step LR
- Polynomial Decay
- Exponential Decay
- Regularization
- Weight Decay
- Label Smoothing
- Stochastic Depth (depends on networks)
- Dropout (depends on networks)
- Loss
- Cross Entropy (w/ class weight and auxiliary logit support)
- Binary Cross Entropy (w/ class weight and auxiliary logit support)
- Soft Cross Entropy Loss (automatically enabled if mixup or label smoothing is used)
- Soft Binary Cross Entropy Loss (automatically enabled if mixup or label smoothing is used)
- Ensemble
- Warmup EMA (Exponential Moving Average)
- 2024/1/17
Release 0.3.0 is published. We will drop MindSpore 1.x in the future release.
- New models:
- Features:
- AsymmetricLoss & JSDCrossEntropy
- Augmentations Split
- Customized AMP
- Bug fixes:
- Since the classifier weights are not fully deleted, you may encounter an error passing in the
num_classeswhen creating a pre-trained model.
- Since the classifier weights are not fully deleted, you may encounter an error passing in the
- Refactoring:
- Documentation:
- A guide of how to extract multiscale features from backbone.
- A guide of how to finetune the pre-trained model on a custom dataset.
- BREAKING CHANGES:
- We are going to drop support of MindSpore 1.x for it's EOL.
- Configuration
filter_bias_and_bnwill be removed and renamed asweight_decay_filter, due to a prolonged misunderstanding of the MindSpore optimizer. We will migrate the existing training recipes, but the signature change of functioncreate_optimizerwill be incompatible and the old version training recipes will also be incompatible. See PR/752 for details.
See RELEASE for detailed history.
We appreciate all kinds of contributions including issues and PRs to make MindCV better.
Please refer to CONTRIBUTING.md for the contributing guideline. Please follow the Model Template and Guideline for contributing a model that fits the overall interface :)
This project follows the Apache License 2.0 open-source license.
MindCV is an open-source project jointly developed by the MindSpore team, Xidian University, and Xi'an Jiaotong University. Sincere thanks to all participating researchers and developers for their hard work on this project. We also acknowledge the computing resources provided by OpenI.
If you find this project useful in your research, please consider citing:
@misc{MindSpore Computer Vision 2022,
title={{MindSpore Computer Vision}:MindSpore Computer Vision Toolbox and Benchmark},
author={MindSpore Vision Contributors},
howpublished = {\url{https://github.com/mindspore-lab/mindcv/}},
year={2022}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent