该内容围绕蛇类识别模型展开，先安装PaddleX，解压数据集并划分，设置GPU后进行图像预处理与增强，定义数据迭代器，用ResNet50_vd_ssld模型训练，接着导出并转换模型为PaddleHub模块，测试单张和多张图片识别效果，最后介绍在GitHub提pr的步骤。
☞☞☞AI 智能聊天, 问答助手, AI 智能搜索, 免费无限量使用 DeepSeek R1 模型☜☜☜
一、模型开发
1.安装必要的资源库
原项目使用PaddleX开发，因此这里先安装PaddleX：
In [5]
!pip install paddlex
2.数据预处理
2.1解压数据集
In [2]
!unzip data/data44587/snake_data.zip -d /home/aistudio/
2.2划分训练集
In [ ]
!paddlex --split_dataset --format ImageNet --dataset_dir '/home/aistudio/snake_data' --val_value 0.2 --test_value 0.1
3.模型训练
3.1设置使用0号GPU卡
In [ ]
import matplotlib
matplotlib.use('Agg')
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'import paddlex as pdx
3.2图像预处理+数据增强
In [ ]
from paddlex.cls import transforms
train_transforms = transforms.Compose([
transforms.RandomCrop(crop_size=224),
transforms.RandomHorizontalFlip(),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.ResizeByShort(short_size=256),
transforms.CenterCrop(crop_size=224),
transforms.Normalize()
])
3.3数据迭代器的定义
In [ ]
train_dataset = pdx.datasets.ImageNet(
data_dir='snake_data',
file_list='snake_data/train_list.txt',
label_list='snake_data/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.ImageNet(
data_dir='snake_data',
file_list='snake_data/val_list.txt',
label_list='snake_data/labels.txt',
transforms=eval_transforms)
2020-07-19 11:49:17 [INFO] Starting to read file list from dataset... 2020-07-19 11:49:17 [INFO] 17364 samples in file snake_data/train_list.txt 2020-07-19 11:49:17 [INFO] Starting to read file list from dataset... 2020-07-19 11:49:17 [INFO] 25 samples in file snake_data/val_list.txt
3.4开始炼丹
In [ ]
num_classes = len(train_dataset.labels)
model = pdx.cls.ResNet50_vd_ssld(num_classes=num_classes)
model.train(num_epochs = 60,
save_interval_epochs = 10,
train_dataset = train_dataset,
train_batch_size = 64,
eval_dataset = eval_dataset,
learning_rate = 0.025,
warmup_steps = 1084,
warmup_start_lr = 0.0001,
lr_decay_epochs=[20, 40],
lr_decay_gamma = 0.025,
save_dir='/home/aistudio',
use_vdl=True)
4.查看模型预测效果
In [ ]
import cv2import matplotlib.pyplot as plt# 加载模型print('**************************************加载模型*****************************************')
model = pdx.load_model('best_model')# 显示图片img = cv2.imread('test.jpg')
b,g,r = cv2.split(img)
img = cv2.merge([r,g,b])
%matplotlib inline
plt.imshow(img)# 预测result = model.predict('test.jpg', topk=3)print('**************************************预测*****************************************')print(result[0])
**************************************加载模型*****************************************
2020-07-19 14:21:06 [INFO] Model[ResNet50_vd_ssld] loaded.
**************************************预测*****************************************
{'category_id': 4, 'category': '西部菱斑响尾蛇', 'score': 0.9999999}
<Figure size 432x288 with 1 Axes>
二、封装Module
1.导出inference模型
| 参数 | 说明 |
|---|---|
| --model_dir | inference模型所在的文件地址，文件包括：.pdparams、.pdopt、.pdmodel、.json和.yml |
| --save_dir | 导出inference模型，文件将包括：__model__、__params__和model.yml |
In [ ]
!paddlex --export_inference --model_dir=best_model --save_dir=./inference_model/ResNet50_vd_ssld
W0717 23:24:19.157521 13809 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 9.2, Runtime API Version: 9.0 W0717 23:24:19.161340 13809 device_context.cc:260] device: 0, cuDNN Version: 7.3. 2020-07-17 23:24:22 [INFO] Model[ResNet50_vd_ssld] loaded. 2020-07-17 23:24:22 [INFO] Model for inference deploy saved in ./inference_model/ResNet50_vd_ssld.
2.模型转换
PaddleX模型可以快速转换成PaddleHub模型，只需要用下面这一句命令即可：
In [1]
!hub convert --model_dir inference_model/ResNet50_vd_ssld \
--module_name SnakeIdentification \
--module_version 1.0.0 \
--output_dir outputs
转换成功后的模型保存在outputs文件夹下，我们解压一下：
In [3]
!gzip -dfq /home/aistudio/outputs/SnakeIdentification.tar.gz !tar -xf /home/aistudio/outputs/SnakeIdentification.tar
3.模型安装
安装我们刚刚转换的模型：
In [6]
!hub install SnakeIdentification
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/setuptools/depends.py:2: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses import imp /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/__init__.py:107: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working from collections import MutableMapping /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/rcsetup.py:20: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working from collections import Iterable, Mapping /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/colors.py:53: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working from collections import Sized[2021-03-12 10:45:54,828] [ INFO] - Successfully uninstalled SnakeIdentification[2021-03-12 10:45:55,105] [ INFO] - Successfully installed SnakeIdentification-1.0.0
4.模型预测
预测单张图片
In [12]
import cv2import paddlehub as hub
module = hub.Module(name="SnakeIdentification")
images = [cv2.imread('snake_data/class_1/2421.jpg')]# execute predict and print the resultresults = module.predict(images=images)for result in results: print(result)
[2021-03-12 10:55:05,972] [ WARNING] - The _initialize method in HubModule will soon be deprecated, you can use the __init__() to handle the initialization of the object
[{'category_id': 0, 'category': '水蛇', 'score': 0.9999205}]
预测多张图片
选取5张图片，每张图片对应一个类别：
In [13]
import cv2import paddlehub as hub
module = hub.Module(name="SnakeIdentification")
images = [cv2.imread('snake_data/class_1/2421.jpg'), cv2.imread('snake_data/class_2/113.jpg'), cv2.imread('snake_data/class_3/757.jpg'),
cv2.imread('snake_data/class_4/1101.jpg'), cv2.imread('snake_data/class_5/2566.jpg')]# execute predict and print the resultresults = module.predict(images=images)for result in results: print(result)
[2021-03-12 11:00:07,036] [ WARNING] - The _initialize method in HubModule will soon be deprecated, you can use the __init__() to handle the initialization of the object
[{'category_id': 0, 'category': '水蛇', 'score': 0.9999205}]
[{'category_id': 1, 'category': '剑纹带蛇', 'score': 0.9988399}]
[{'category_id': 2, 'category': '德凯斯氏蛇', 'score': 0.9867851}]
[{'category_id': 3, 'category': '黑鼠蛇', 'score': 0.9468411}]
[{'category_id': 4, 'category': '西部菱斑响尾蛇', 'score': 1.0}]
三、在GitHub上提pr
pr就是Pull Request（翻译过来就是：拉取请求）的简称
2.上传Module
本项目是图像分类的项目，所以进入到图像分类的目录下：
PaddleHub/modules/image/classification/
点击Add file：
先输入您上传的Module名称，这里我的Module名称命名为SnakeIdentification，将它变成一个文件夹，只需要在后面加一个‘/’,创建好文件夹以后，把Module里的文件上传上去即可：
上传成功后，点击Commit，文件就会自动上传到你自己的代码仓库里
3.Pull Request
最后一步，拉取请求：
确认无误后点击提交即可：
