Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
langdu 7aad3c8feb | 2 years ago | |
---|---|---|
.. | ||
_base_ | 2 years ago | |
ann | 2 years ago | |
attention_unet | 2 years ago | |
bisenet | 2 years ago | |
bisenetv1 | 2 years ago | |
danet | 2 years ago | |
decoupled_segnet | 2 years ago | |
deeplabv3 | 2 years ago | |
deeplabv3p | 2 years ago | |
dnlnet | 2 years ago | |
emanet | 2 years ago | |
fastscnn | 2 years ago | |
fcn | 2 years ago | |
gcnet | 2 years ago | |
ginet | 2 years ago | |
gscnn | 2 years ago | |
hardnet | 2 years ago | |
isanet | 2 years ago | |
ocrnet | 2 years ago | |
pointrend | 2 years ago | |
portraitnet | 2 years ago | |
pp_humanseg_lite | 2 years ago | |
pspnet | 2 years ago | |
quick_start | 2 years ago | |
segformer | 2 years ago | |
segnet | 2 years ago | |
setr | 2 years ago | |
sfnet | 2 years ago | |
stdcseg | 2 years ago | |
u2net | 2 years ago | |
unet | 2 years ago | |
unet_3plus | 2 years ago | |
unet_plusplus | 2 years ago | |
README.md | 2 years ago |
训练数据集
- 参数
- type : 数据集类型,所支持值请参考训练配置文件
- others : 请参考对应模型训练配置文件
评估数据集
- 参数
- type : 数据集类型,所支持值请参考训练配置文件
- others : 请参考对应模型训练配置文件
单张卡上,每步迭代训练时的数据量
训练步数
训练优化器
- 参数
- type : 优化器类型,支持目前Paddle官方所有优化器
- weight_decay : L2正则化的值
- others : 请参考Paddle官方Optimizer文档
学习率
- 参数
- type : 学习率类型,支持10种策略,分别是'PolynomialDecay', 'PiecewiseDecay', 'StepDecay', 'CosineAnnealingDecay', 'ExponentialDecay', 'InverseTimeDecay', 'LinearWarmup', 'MultiStepDecay', 'NaturalExpDecay', 'NoamDecay'.
- others : 请参考Paddle官方LRScheduler文档
lr_scheduler
代替)学习率
- 参数
- value : 初始学习率
- decay : 衰减配置
- type : 衰减类型,目前只支持poly
- power : 衰减率
- end_lr : 最终学习率
损失函数
- 参数
- types : 损失函数列表
- type : 损失函数类型,所支持值请参考损失函数库
- ignore_index : 训练过程需要忽略的类别,默认取值与
train_dataset
的ignore_index一致,推荐不用设置此项。如果设置了此项,loss
和train_dataset
的ignore_index必须相同。- coef : 对应损失函数列表的系数列表
待训练模型
- 参数
- type : 模型类型,所支持值请参考模型库
- others : 请参考对应模型训练配置文件
模型导出配置
- 参数
- transforms : 预测时的预处理操作,支持配置的transforms与
train_dataset
、val_dataset
等相同。如果不填写该项,默认只会对数据进行归一化标准化操作。
具体配置文件说明请参照配置文件详解
使用Paddle复现BisNetV1
Python Java Shell Text Gradle other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》