forked from mindspore-Ecosystem/mindspore
448 lines
16 KiB
ReStructuredText
448 lines
16 KiB
ReStructuredText
mindspore.dataset.transforms
|
||
============================
|
||
|
||
通用
|
||
-----
|
||
|
||
此模块用于通用数据增强，其中一部分增强操作是用C++实现的，具有较好的高性能，另一部分是基于Python实现，使用了NumPy模块作为支持。
|
||
|
||
在API示例中，常用的模块导入方法如下：
|
||
|
||
.. code-block::
|
||
|
||
import mindspore.dataset as ds
|
||
import mindspore.dataset.transforms as transforms
|
||
|
||
注意：旧的API导入方式已经过时且会逐步废弃，因此推荐使用上面的方式，但目前仍可按以下方式导入：
|
||
|
||
.. code-block::
|
||
|
||
from mindspore.dataset.transforms import c_transforms
|
||
from mindspore.dataset.transforms import py_transforms
|
||
|
||
更多详情请参考 `通用数据变换 <https://www.mindspore.cn/tutorials/zh-CN/master/beginner/transforms.html#common-transforms>`_ 。
|
||
|
||
常用数据处理术语说明如下：
|
||
|
||
- TensorOperation，所有C++实现的数据处理操作的基类。
|
||
- PyTensorOperation，所有Python实现的数据处理操作的基类。
|
||
|
||
变换
|
||
^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_transforms
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.transforms.Compose
|
||
mindspore.dataset.transforms.Concatenate
|
||
mindspore.dataset.transforms.Duplicate
|
||
mindspore.dataset.transforms.Fill
|
||
mindspore.dataset.transforms.Mask
|
||
mindspore.dataset.transforms.OneHot
|
||
mindspore.dataset.transforms.PadEnd
|
||
mindspore.dataset.transforms.RandomApply
|
||
mindspore.dataset.transforms.RandomChoice
|
||
mindspore.dataset.transforms.RandomOrder
|
||
mindspore.dataset.transforms.Slice
|
||
mindspore.dataset.transforms.TypeCast
|
||
mindspore.dataset.transforms.Unique
|
||
|
||
工具
|
||
^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_transforms
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.transforms.Relational
|
||
|
||
视觉
|
||
-----
|
||
|
||
此模块用于图像数据增强，其中有一部分增强是基于C++ OpenCV实现的，具有较好的性能，而另一部分是基于Python Pillow实现的。
|
||
|
||
API样例中常用的导入模块如下：
|
||
|
||
.. code-block::
|
||
|
||
import mindspore.dataset as ds
|
||
import mindspore.dataset.vision as vision
|
||
import mindspore.dataset.vision.utils as utils
|
||
|
||
注意：旧的API导入方式已经过时且会逐步废弃，因此推荐使用上面的方式，但目前仍可按以下方式导入：
|
||
|
||
.. code-block::
|
||
|
||
import mindspore.dataset.vision.c_transforms as c_vision
|
||
import mindspore.dataset.vision.py_transforms as py_vision
|
||
from mindspore.dataset.transforms import c_transforms
|
||
|
||
更多详情请参考 `视觉数据变换 <https://www.mindspore.cn/tutorials/zh-CN/master/beginner/transforms.html#vision-transforms>`_ 。
|
||
|
||
常用数据处理术语说明如下：
|
||
|
||
- TensorOperation，所有C++实现的数据处理操作的基类。
|
||
- PyTensorOperation，所有Python实现的数据处理操作的基类。
|
||
|
||
数据增强操作可以放入数据处理Pipeline中执行，也可以Eager模式执行：
|
||
|
||
- Pipeline模式一般用于处理数据集，示例可参考 `数据处理Pipeline介绍 <https://www.mindspore.cn/docs/zh-CN/master/api_python/mindspore.dataset.html#数据处理pipeline介绍>`_ 。
|
||
- Eager模式一般用于零散样本，图像预处理举例如下：
|
||
|
||
.. code-block::
|
||
|
||
import numpy as np
|
||
import mindspore.dataset.vision as vision
|
||
from PIL import Image,ImageFont,ImageDraw
|
||
|
||
# 画圆形
|
||
img = Image.new("RGB", (300, 300), (255, 255, 255))
|
||
draw = ImageDraw.Draw(img)
|
||
draw.ellipse(((0, 0), (100, 100)), fill=(255, 0, 0), outline=(255, 0, 0), width=5)
|
||
img.save("./1.jpg")
|
||
with open("./1.jpg", "rb") as f:
|
||
data = f.read()
|
||
|
||
data_decoded = vision.Decode()(data)
|
||
data_croped = vision.RandomCrop(size=(250, 250))(data_decoded)
|
||
data_resized = vision.Resize(size=(224, 224))(data_croped)
|
||
data_normalized = vision.Normalize(mean=[0.485 * 255, 0.456 * 255, 0.406 * 255],
|
||
std=[0.229 * 255, 0.224 * 255, 0.225 * 255])(data_resized)
|
||
data_hwc2chw = vision.HWC2CHW()(data_normalized)
|
||
print("data: {}, shape: {}".format(data_hwc2chw, data_hwc2chw.shape), flush=True)
|
||
|
||
变换
|
||
^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_vision
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.vision.AdjustBrightness
|
||
mindspore.dataset.vision.AdjustContrast
|
||
mindspore.dataset.vision.AdjustGamma
|
||
mindspore.dataset.vision.AdjustHue
|
||
mindspore.dataset.vision.AdjustSaturation
|
||
mindspore.dataset.vision.AdjustSharpness
|
||
mindspore.dataset.vision.Affine
|
||
mindspore.dataset.vision.AutoAugment
|
||
mindspore.dataset.vision.AutoContrast
|
||
mindspore.dataset.vision.BoundingBoxAugment
|
||
mindspore.dataset.vision.CenterCrop
|
||
mindspore.dataset.vision.ConvertColor
|
||
mindspore.dataset.vision.Crop
|
||
mindspore.dataset.vision.CutMixBatch
|
||
mindspore.dataset.vision.CutOut
|
||
mindspore.dataset.vision.Decode
|
||
mindspore.dataset.vision.Equalize
|
||
mindspore.dataset.vision.Erase
|
||
mindspore.dataset.vision.FiveCrop
|
||
mindspore.dataset.vision.GaussianBlur
|
||
mindspore.dataset.vision.Grayscale
|
||
mindspore.dataset.vision.HorizontalFlip
|
||
mindspore.dataset.vision.HsvToRgb
|
||
mindspore.dataset.vision.HWC2CHW
|
||
mindspore.dataset.vision.Invert
|
||
mindspore.dataset.vision.LinearTransformation
|
||
mindspore.dataset.vision.MixUp
|
||
mindspore.dataset.vision.MixUpBatch
|
||
mindspore.dataset.vision.Normalize
|
||
mindspore.dataset.vision.NormalizePad
|
||
mindspore.dataset.vision.Pad
|
||
mindspore.dataset.vision.PadToSize
|
||
mindspore.dataset.vision.Perspective
|
||
mindspore.dataset.vision.Posterize
|
||
mindspore.dataset.vision.RandAugment
|
||
mindspore.dataset.vision.RandomAdjustSharpness
|
||
mindspore.dataset.vision.RandomAffine
|
||
mindspore.dataset.vision.RandomAutoContrast
|
||
mindspore.dataset.vision.RandomColor
|
||
mindspore.dataset.vision.RandomColorAdjust
|
||
mindspore.dataset.vision.RandomCrop
|
||
mindspore.dataset.vision.RandomCropDecodeResize
|
||
mindspore.dataset.vision.RandomCropWithBBox
|
||
mindspore.dataset.vision.RandomEqualize
|
||
mindspore.dataset.vision.RandomErasing
|
||
mindspore.dataset.vision.RandomGrayscale
|
||
mindspore.dataset.vision.RandomHorizontalFlip
|
||
mindspore.dataset.vision.RandomHorizontalFlipWithBBox
|
||
mindspore.dataset.vision.RandomInvert
|
||
mindspore.dataset.vision.RandomLighting
|
||
mindspore.dataset.vision.RandomPerspective
|
||
mindspore.dataset.vision.RandomPosterize
|
||
mindspore.dataset.vision.RandomResizedCrop
|
||
mindspore.dataset.vision.RandomResizedCropWithBBox
|
||
mindspore.dataset.vision.RandomResize
|
||
mindspore.dataset.vision.RandomResizeWithBBox
|
||
mindspore.dataset.vision.RandomRotation
|
||
mindspore.dataset.vision.RandomSelectSubpolicy
|
||
mindspore.dataset.vision.RandomSharpness
|
||
mindspore.dataset.vision.RandomSolarize
|
||
mindspore.dataset.vision.RandomVerticalFlip
|
||
mindspore.dataset.vision.RandomVerticalFlipWithBBox
|
||
mindspore.dataset.vision.Rescale
|
||
mindspore.dataset.vision.Resize
|
||
mindspore.dataset.vision.ResizedCrop
|
||
mindspore.dataset.vision.ResizeWithBBox
|
||
mindspore.dataset.vision.RgbToHsv
|
||
mindspore.dataset.vision.Rotate
|
||
mindspore.dataset.vision.SlicePatches
|
||
mindspore.dataset.vision.Solarize
|
||
mindspore.dataset.vision.TenCrop
|
||
mindspore.dataset.vision.ToNumpy
|
||
mindspore.dataset.vision.ToPIL
|
||
mindspore.dataset.vision.ToTensor
|
||
mindspore.dataset.vision.ToType
|
||
mindspore.dataset.vision.TrivialAugmentWide
|
||
mindspore.dataset.vision.UniformAugment
|
||
mindspore.dataset.vision.VerticalFlip
|
||
|
||
工具
|
||
^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_vision
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.vision.AutoAugmentPolicy
|
||
mindspore.dataset.vision.Border
|
||
mindspore.dataset.vision.ConvertMode
|
||
mindspore.dataset.vision.ImageBatchFormat
|
||
mindspore.dataset.vision.ImageReadMode
|
||
mindspore.dataset.vision.Inter
|
||
mindspore.dataset.vision.SliceMode
|
||
mindspore.dataset.vision.encode_jpeg
|
||
mindspore.dataset.vision.encode_png
|
||
mindspore.dataset.vision.get_image_num_channels
|
||
mindspore.dataset.vision.get_image_size
|
||
mindspore.dataset.vision.read_file
|
||
mindspore.dataset.vision.read_image
|
||
mindspore.dataset.vision.write_file
|
||
mindspore.dataset.vision.write_jpeg
|
||
mindspore.dataset.vision.write_png
|
||
|
||
文本
|
||
-----
|
||
|
||
此模块用于文本数据增强，包括 `transforms` 和 `utils` 两个子模块。
|
||
|
||
`transforms` 是一个高性能文本数据增强模块，支持常见的文本数据增强处理。
|
||
|
||
`utils` 提供了一些文本处理的工具方法。
|
||
|
||
在API示例中，常用的模块导入方法如下：
|
||
|
||
.. code-block::
|
||
|
||
import mindspore.dataset as ds
|
||
from mindspore.dataset import text
|
||
|
||
更多详情请参考 `文本数据变换 <https://www.mindspore.cn/tutorials/zh-CN/master/beginner/transforms.html#text-transforms>`_ 。
|
||
|
||
常用数据处理术语说明如下：
|
||
|
||
- TensorOperation，所有C++实现的数据处理操作的基类。
|
||
- TextTensorOperation，所有文本数据处理操作的基类，派生自TensorOperation。
|
||
|
||
数据增强操作可以放入数据处理Pipeline中执行，也可以Eager模式执行：
|
||
|
||
- Pipeline模式一般用于处理数据集，示例可参考 `数据处理Pipeline介绍 <https://www.mindspore.cn/docs/zh-CN/master/api_python/mindspore.dataset.html#数据处理pipeline介绍>`_ 。
|
||
- Eager模式一般用于零散样本，文本预处理举例如下：
|
||
|
||
.. code-block::
|
||
|
||
from mindspore.dataset import text
|
||
from mindspore.dataset.text import NormalizeForm
|
||
|
||
# 构造词汇表
|
||
vocab_list = {"床": 1, "前": 2, "明": 3, "月": 4, "光": 5, "疑": 6,
|
||
"是": 7, "地": 8, "上": 9, "霜": 10, "举": 11, "头": 12,
|
||
"望": 13, "低": 14, "思": 15, "故": 16, "乡": 17, "繁": 18,
|
||
"體": 19, "字": 20, "嘿": 21, "哈": 22, "大": 23, "笑": 24,
|
||
"嘻": 25, "UNK": 26}
|
||
vocab = text.Vocab.from_dict(vocab_list)
|
||
tokenizer_op = text.BertTokenizer(vocab=vocab, suffix_indicator='##', max_bytes_per_token=100,
|
||
unknown_token='[UNK]', lower_case=False, keep_whitespace=False,
|
||
normalization_form=NormalizeForm.NONE, preserve_unused_token=True,
|
||
with_offsets=False)
|
||
# 分词
|
||
tokens = tokenizer_op("床前明月光，疑是地上霜，举头望明月，低头思故乡。")
|
||
print("token: {}".format(tokens), flush=True)
|
||
|
||
# 根据单词查找id
|
||
ids = vocab.tokens_to_ids(tokens)
|
||
print("token to id: {}".format(ids), flush=True)
|
||
|
||
# 根据id查找单词
|
||
tokens_from_ids = vocab.ids_to_tokens([15, 3, 7])
|
||
print("token to id: {}".format(tokens_from_ids), flush=True)
|
||
|
||
变换
|
||
^^^^^
|
||
|
||
.. mscnnoteautosummary::
|
||
:toctree: dataset_text
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.text.AddToken
|
||
mindspore.dataset.text.BasicTokenizer
|
||
mindspore.dataset.text.BertTokenizer
|
||
mindspore.dataset.text.CaseFold
|
||
mindspore.dataset.text.FilterWikipediaXML
|
||
mindspore.dataset.text.JiebaTokenizer
|
||
mindspore.dataset.text.Lookup
|
||
mindspore.dataset.text.Ngram
|
||
mindspore.dataset.text.NormalizeUTF8
|
||
mindspore.dataset.text.PythonTokenizer
|
||
mindspore.dataset.text.RegexReplace
|
||
mindspore.dataset.text.RegexTokenizer
|
||
mindspore.dataset.text.SentencePieceTokenizer
|
||
mindspore.dataset.text.SlidingWindow
|
||
mindspore.dataset.text.ToNumber
|
||
mindspore.dataset.text.ToVectors
|
||
mindspore.dataset.text.Truncate
|
||
mindspore.dataset.text.TruncateSequencePair
|
||
mindspore.dataset.text.UnicodeCharTokenizer
|
||
mindspore.dataset.text.UnicodeScriptTokenizer
|
||
mindspore.dataset.text.WhitespaceTokenizer
|
||
mindspore.dataset.text.WordpieceTokenizer
|
||
|
||
|
||
工具
|
||
^^^^^
|
||
|
||
.. mscnnoteautosummary::
|
||
:toctree: dataset_text
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.text.CharNGram
|
||
mindspore.dataset.text.FastText
|
||
mindspore.dataset.text.GloVe
|
||
mindspore.dataset.text.JiebaMode
|
||
mindspore.dataset.text.NormalizeForm
|
||
mindspore.dataset.text.SentencePieceModel
|
||
mindspore.dataset.text.SentencePieceVocab
|
||
mindspore.dataset.text.SPieceTokenizerLoadType
|
||
mindspore.dataset.text.SPieceTokenizerOutType
|
||
mindspore.dataset.text.Vectors
|
||
mindspore.dataset.text.Vocab
|
||
mindspore.dataset.text.to_bytes
|
||
mindspore.dataset.text.to_str
|
||
|
||
音频
|
||
-----
|
||
|
||
.. include:: dataset_audio/mindspore.dataset.audio.rst
|
||
|
||
数据增强操作可以放入数据处理Pipeline中执行，也可以Eager模式执行：
|
||
|
||
- Pipeline模式一般用于处理数据集，示例可参考 `数据处理Pipeline介绍 <https://www.mindspore.cn/docs/zh-CN/master/api_python/mindspore.dataset.html#数据处理pipeline介绍>`_ 。
|
||
- Eager模式一般用于零散样本，音频预处理举例如下：
|
||
|
||
.. code-block::
|
||
|
||
import numpy as np
|
||
import mindspore.dataset.audio as audio
|
||
from mindspore.dataset.audio import ResampleMethod
|
||
|
||
# 音频输入
|
||
waveform = np.random.random([1, 30])
|
||
|
||
# 增强操作
|
||
resample_op = audio.Resample(orig_freq=48000, new_freq=16000,
|
||
resample_method=ResampleMethod.SINC_INTERPOLATION,
|
||
lowpass_filter_width=6, rolloff=0.99, beta=None)
|
||
waveform_resampled = resample_op(waveform)
|
||
print("waveform reampled: {}".format(waveform_resampled), flush=True)
|
||
|
||
变换
|
||
^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_audio
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.audio.AllpassBiquad
|
||
mindspore.dataset.audio.AmplitudeToDB
|
||
mindspore.dataset.audio.Angle
|
||
mindspore.dataset.audio.BandBiquad
|
||
mindspore.dataset.audio.BandpassBiquad
|
||
mindspore.dataset.audio.BandrejectBiquad
|
||
mindspore.dataset.audio.BassBiquad
|
||
mindspore.dataset.audio.Biquad
|
||
mindspore.dataset.audio.ComplexNorm
|
||
mindspore.dataset.audio.ComputeDeltas
|
||
mindspore.dataset.audio.Contrast
|
||
mindspore.dataset.audio.DBToAmplitude
|
||
mindspore.dataset.audio.DCShift
|
||
mindspore.dataset.audio.DeemphBiquad
|
||
mindspore.dataset.audio.DetectPitchFrequency
|
||
mindspore.dataset.audio.Dither
|
||
mindspore.dataset.audio.EqualizerBiquad
|
||
mindspore.dataset.audio.Fade
|
||
mindspore.dataset.audio.Filtfilt
|
||
mindspore.dataset.audio.Flanger
|
||
mindspore.dataset.audio.FrequencyMasking
|
||
mindspore.dataset.audio.Gain
|
||
mindspore.dataset.audio.GriffinLim
|
||
mindspore.dataset.audio.HighpassBiquad
|
||
mindspore.dataset.audio.InverseMelScale
|
||
mindspore.dataset.audio.InverseSpectrogram
|
||
mindspore.dataset.audio.LFCC
|
||
mindspore.dataset.audio.LFilter
|
||
mindspore.dataset.audio.LowpassBiquad
|
||
mindspore.dataset.audio.Magphase
|
||
mindspore.dataset.audio.MaskAlongAxis
|
||
mindspore.dataset.audio.MaskAlongAxisIID
|
||
mindspore.dataset.audio.MelScale
|
||
mindspore.dataset.audio.MelSpectrogram
|
||
mindspore.dataset.audio.MFCC
|
||
mindspore.dataset.audio.MuLawDecoding
|
||
mindspore.dataset.audio.MuLawEncoding
|
||
mindspore.dataset.audio.Overdrive
|
||
mindspore.dataset.audio.Phaser
|
||
mindspore.dataset.audio.PhaseVocoder
|
||
mindspore.dataset.audio.Resample
|
||
mindspore.dataset.audio.PitchShift
|
||
mindspore.dataset.audio.RiaaBiquad
|
||
mindspore.dataset.audio.SlidingWindowCmn
|
||
mindspore.dataset.audio.SpectralCentroid
|
||
mindspore.dataset.audio.Spectrogram
|
||
mindspore.dataset.audio.TimeMasking
|
||
mindspore.dataset.audio.TimeStretch
|
||
mindspore.dataset.audio.TrebleBiquad
|
||
mindspore.dataset.audio.Vad
|
||
mindspore.dataset.audio.Vol
|
||
|
||
工具
|
||
^^^^^^
|
||
|
||
.. mscnautosummary::
|
||
:toctree: dataset_audio
|
||
:nosignatures:
|
||
:template: classtemplate.rst
|
||
|
||
mindspore.dataset.audio.BorderType
|
||
mindspore.dataset.audio.DensityFunction
|
||
mindspore.dataset.audio.FadeShape
|
||
mindspore.dataset.audio.GainType
|
||
mindspore.dataset.audio.Interpolation
|
||
mindspore.dataset.audio.MelType
|
||
mindspore.dataset.audio.Modulation
|
||
mindspore.dataset.audio.NormMode
|
||
mindspore.dataset.audio.NormType
|
||
mindspore.dataset.audio.ResampleMethod
|
||
mindspore.dataset.audio.ScaleType
|
||
mindspore.dataset.audio.WindowType
|
||
mindspore.dataset.audio.create_dct
|
||
mindspore.dataset.audio.linear_fbanks
|
||
mindspore.dataset.audio.melscale_fbanks
|