Merge branch 'mindspore-lab:main' into swin_trans

Author: wcrzlh

.github/workflows/ci.yml

@@ -34,6 +34,7 @@ jobs:
       run: |
         python -m pip install --upgrade pip
         pip install -r requirements/dev.txt
+        pip install "Pillow==9.1.1"
         # MindSpore must be installed following the instruction from official web, but not from pypi.
         # That's why we exclude mindspore from requirements.txt. Does this work?
         pip install "mindspore>=1.8,<=1.10"

.github/workflows/docs.yml

@@ -21,6 +21,8 @@ jobs:
         run: |
           python -m pip install --upgrade pip
           pip install -r requirements/docs.txt
+          pip install "Pillow==9.1.1"
+          pip install "mindspore>=1.8,<=1.10"
       - name: Build site
         run: mkdocs build
       - name: Deploy to gh-pages

.github/workflows/publish.yml

@@ -24,6 +24,7 @@ jobs:
       run: |
         python -m pip install --upgrade pip
         pip install -r requirements/dev.txt
+        pip install "Pillow==9.1.1"
         pip install "mindspore>=1.8,<=1.10"
         pip install build twine
     - name: Build package

README.md

@@ -13,21 +13,22 @@
 [![Imports: isort](https://img.shields.io/badge/%20imports-isort-%231674b1?style=flat&labelColor=ef8336)](https://pycqa.github.io/isort/)
 [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit)](https://github.com/pre-commit/pre-commit)
 
-English | [中文](README_CN.md)
+[📝Documentation](https://mindspore-lab.github.io/mindcv/) |
+[🚀Installation](https://mindspore-lab.github.io/mindcv/installation/) |
+[🎁Model Zoo](https://mindspore-lab.github.io/mindcv/modelzoo/) |
+[🎉Update News](https://github.com/mindspore-lab/mindcv/blob/main/RELEASE.md) |
+[🐛Reporting Issues](https://github.com/mindspore-lab/mindcv/issues/new/choose)
 
-[Introduction](#introduction) |
-[Installation](#installation) |
-[Get Started](#get-started) |
-[Tutorials](#tutorials) |
-[Model List](#model-list) |
-[Supported Algorithms](#supported-algorithms)
+English | [中文](README_CN.md)
 
 </div>
 
 ## Introduction
 
 MindCV is an open-source toolbox for computer vision research and development based on [MindSpore](https://www.mindspore.cn/en). 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 main branch works with **MindSpore 1.8+**, including **MindSpore 2.0🔥**.
+
 ### Major Features
 
 - **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:
@@ -76,9 +77,11 @@ Below are a few code snippets for your taste.
 ['swin_tiny']
 # Create the model object
 >>> network = mindcv.create_model('swin_tiny', pretrained=True)
+```
+```shell
 # Validate its accuracy
->>> !python validate.py --model=swin_tiny --pretrained --dataset=imagenet --val_split=validation
-{'Top_1_Accuracy': 0.808343989769821, 'Top_5_Accuracy': 0.9527253836317136, 'loss': 0.8474242982580839}
+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**
@@ -91,9 +94,9 @@ Right click on the image below and save as `dog.jpg`.
 
 Classify the downloaded image with a pretrained SoTA model:
 
-```pycon
->>> !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}
+```shell
+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.
 
@@ -292,9 +295,22 @@ Please see [configs](./configs) for the details about model performance and pret
 
 ## What is New
 
-- 2023/6/2
-1. New version: `0.2.1` is released!
-2. New [documents](https://mindspore-lab.github.io/mindcv/) is online!
+- 2023/6/16
+1. New version `0.2.2` is released! We upgrade to support `MindSpore` v2.0 while maintaining compatibility of v1.8
+2. New models:
+   - [ConvNextV2](configs/convnextv2)
+   - mini of [CoAT](configs/coat)
+   - 1.3 of [MnasNet](configs/mnasnet)
+   - AMP(O3) version of [ShuffleNetV2](configs/shufflenetv2)
+3. New features:
+   - Gradient Accumulation
+   - DynamicLossScale for customized [TrainStep](mindcv/utils/train_step.py)
+   - OneCycleLR and CyclicLR learning rate scheduler
+   - Refactored Logging
+   - Pyramid Feature Extraction
+4. Bug fixes:
+   - Serving Deployment Tutorial(mobilenet_v3 doesn't work on ms1.8 when using Ascend backend)
+   - Some broken links on our documentation website.
 
 See [RELEASE](RELEASE.md) for detailed history.
 

README_CN.md

@@ -13,21 +13,22 @@
 [![Imports: isort](https://img.shields.io/badge/%20imports-isort-%231674b1?style=flat&labelColor=ef8336)](https://pycqa.github.io/isort/)
 [![pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit)](https://github.com/pre-commit/pre-commit)
 
-[English](README.md) | 中文
+[📝使用文档](https://mindspore-lab.github.io/mindcv/zh/) |
+[🚀安装教程](https://mindspore-lab.github.io/mindcv/zh/installation/) |
+[🎁模型仓库](https://mindspore-lab.github.io/mindcv/zh/modelzoo/) |
+[🎉更新日志](https://github.com/mindspore-lab/mindcv/blob/main/RELEASE.md) |
+[🐛报告问题](https://github.com/mindspore-lab/mindcv/issues/new/choose)
 
-[简介](#简介) |
-[安装](#安装) |
-[快速入门](#快速入门) |
-[教程](#教程) |
-[模型列表](#模型列表) |
-[支持算法](#支持算法)
+[English](README.md) | 中文
 
 </div>
 
 ## 简介
 
 MindCV是一个基于 [MindSpore](https://www.mindspore.cn/) 开发的，致力于计算机视觉相关技术研发的开源工具箱。它提供大量的计算机视觉领域的经典模型和SoTA模型以及它们的预训练权重和训练策略。同时，还提供了自动增强等SoTA算法来提高模型性能。通过解耦的模块设计，您可以轻松地将MindCV应用到您自己的CV任务中。
 
+主分支代码目前支持 **MindSpore 1.8+** 以上的版本，包含 **MindSpore 2.0🔥** 版本。
+
 ### 主要特性
 
 - **高易用性** MindCV将视觉任务分解为各种可配置的组件，用户可以轻松地构建自己的数据处理和模型训练流程。
@@ -76,9 +77,12 @@ MindCV是一个基于 [MindSpore](https://www.mindspore.cn/) 开发的，致力
 ['swin_tiny']
 # 创建模型
 >>> network = mindcv.create_model('swin_tiny', pretrained=True)
-# 验证模型的准确率
->>> !python validate.py --model=swin_tiny --pretrained --dataset=imagenet --val_split=validation
-{'Top_1_Accuracy': 0.808343989769821, 'Top_5_Accuracy': 0.9527253836317136, 'loss': 0.8474242982580839}
+```
+
+```shell
+# 验证模型的准确度
+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}
 ```
 
 **图片分类示例**
@@ -91,9 +95,9 @@ MindCV是一个基于 [MindSpore](https://www.mindspore.cn/) 开发的，致力
 
 使用加载了预训练参数的SoTA模型对图片进行推理。
 
-```pycon
->>> !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}
+```shell
+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}
 ```
 
 预测结果排名前1的是拉布拉多犬，正是这张图片里的狗狗的品种。
@@ -294,6 +298,23 @@ python train.py --model=resnet50 --dataset=cifar10 \
 
 ## 更新
 
+- 2023/6/16
+1. 新版本 `0.2.2` 发布啦！我们将`MindSpore`升级到了2.0版本，同时保持了对1.8版本的兼容
+2. 新模型:
+   - [ConvNextV2](configs/convnextv2)
+   - [CoAT](configs/coat)的mini规格
+   - [MnasNet](configs/mnasnet)的1.3规格
+   - [ShuffleNetV2](configs/shufflenetv2)的混合精度(O3)版本
+3. 新特性:
+   - 梯度累加
+   - 自定义[TrainStep](mindcv/utils/train_step.py)支持了动态损失缩放
+   - `OneCycleLR`和`CyclicLR`学习率调度器
+   - 更好的日志打印与记录
+   - 金字塔特征抽取
+4. 错误修复:
+   - `Serving`部署教程(mobilenet_v3在昇腾后端的MindSpore1.8版本上不支持)
+   - 文档网站上的损坏链接
+
 - 2023/6/2
 1. 新版本：`0.2.1` 发布
 2. 新[文档](https://mindspore-lab.github.io/mindcv/zh/)上线

RELEASE.md

@@ -1,5 +1,23 @@
 # Release Note
 
+## 0.2.2 (2023/6/16)
+
+1. New version `0.2.2` is released! We upgrade to support `MindSpore` v2.0 while maintaining compatibility of v1.8
+2. New models:
+   - [ConvNextV2](configs/convnextv2)
+   - mini of [CoAT](configs/coat)
+   - 1.3 of [MnasNet](configs/mnasnet)
+   - AMP(O3) version of [ShuffleNetV2](configs/shufflenetv2)
+3. New features:
+   - Gradient Accumulation
+   - DynamicLossScale for customized [TrainStep](mindcv/utils/train_step.py)
+   - OneCycleLR and CyclicLR learning rate scheduler
+   - Refactored Logging
+   - Pyramid Feature Extraction
+4. Bug fixes:
+   - Serving Deployment Tutorial(mobilenet_v3 doesn't work on ms1.8 when using Ascend backend)
+   - Some broken links on our documentation website.
+
 ## 0.2.1
 
 - 2023/6/2

config.py

@@ -232,6 +232,11 @@ def create_parser():
                             'Choice: O0 - all FP32, O1 - only cast ops in white-list to FP16, '
                             'O2 - cast all ops except for blacklist to FP16, '
                             'O3 - cast all ops to FP16. (default="O0").')
+    group.add_argument('--val_amp_level', type=str, default='O0',
+                       help='Amp level used in validation '
+                            'Choice: O0 - all FP32, O1 - only cast ops in white-list to FP16, '
+                            'O2 - cast all ops except for blacklist to FP16, '
+                            'O3 - cast all ops to FP16. (default="O0").')
     group.add_argument('--loss_scale_type', type=str, default='fixed',
                        choices=['fixed', 'dynamic', 'auto'],
                        help='The type of loss scale (default="fixed")')

configs/bit/README.md

@@ -81,7 +81,7 @@ python validate.py -c configs/bit/bit_resnet50_ascend.yaml --data_dir /path/to/i
 
 ### Deployment
 
-Please refer to the [deployment tutorial](https://github.com/mindspore-lab/mindcv/blob/main/tutorials/deployment.md) in MindCV.
+Please refer to the [deployment tutorial](https://mindspore-lab.github.io/mindcv/tutorials/deployment/) in MindCV.
 
 ## References
 

configs/coat/README.md

@@ -74,7 +74,7 @@ python validate.py -c configs/coat/coat_lite_tiny_ascend.yaml --data_dir /path/t
 
 ### Deployment
 
-To deploy online inference services with the trained model efficiently, please refer to the [deployment tutorial](https://github.com/mindspore-lab/mindcv/blob/main/tutorials/deployment.md).
+To deploy online inference services with the trained model efficiently, please refer to the [deployment tutorial](https://mindspore-lab.github.io/mindcv/tutorials/deployment/).
 
 ## References
 

configs/coat/coat_mini_ascend.yaml

@@ -7,7 +7,7 @@ val_interval: 1
 
 # dataset
 dataset: 'imagenet'
-data_dir: '/path/to/imagenet/'
+data_dir: '/path/to/imagenet'
 shuffle: True
 dataset_download: False
 batch_size: 32
@@ -59,3 +59,4 @@ filter_bias_and_bn: True
 loss_scale: 4096
 use_nesterov: False
 loss_scale_type: dynamic
+drop_overflow_update: True