Xc@@sdZddlmZddlmZddlZddlZddlmZddl j Z ddl m Z ddlZddlZddlZddlmZydd lmZWnek reZnXd ddd d d Zd ddd d dZd ddd d dZd ddd d dZddZdZdd d dZdZ ee!dZ"edZ#e$edZ%ddZ&de'fdYZ(de'fdYZ)de)fdYZ*d e)fd!YZ+dS("sFairly basic set of tools for real-time data augmentation on image data. Can easily be extended to include new transformations, new preprocessing methods, etc... i(tabsolute_import(tprint_functionN(tlinalg(trangei(tbackend(tImageitnearestgc C@stjdtjj| |}tjtj|tj| dgtj|tj|dgdddgg}|j||j|} } t|| | } t || |||}|S(sPerforms a random rotation of a Numpy image tensor. # Arguments x: Input tensor. Must be 3D. rg: Rotation range, in degrees. row_axis: Index of axis for rows in the input tensor. col_axis: Index of axis for columns in the input tensor. channel_axis: Index of axis for channels in the input tensor. fill_mode: Points outside the boundaries of the input are filled according to the given mode (one of `{'constant', 'nearest', 'reflect', 'wrap'}`). cval: Value used for points outside the boundaries of the input if `mode='constant'`. # Returns Rotated Numpy image tensor. iii( tnptpitrandomtuniformtarraytcostsintshapettransform_matrix_offset_centertapply_transform( txtrgtrow_axistcol_axist channel_axist fill_modetcvaltthetatrotation_matrixthtwttransform_matrix((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytrandom_rotations!%cC@s|j||j|}} tjj| ||} tjj| || } tjdd| gdd| gdddgg} | } t|| |||}|S(sPerforms a random spatial shift of a Numpy image tensor. # Arguments x: Input tensor. Must be 3D. wrg: Width shift range, as a float fraction of the width. hrg: Height shift range, as a float fraction of the height. row_axis: Index of axis for rows in the input tensor. col_axis: Index of axis for columns in the input tensor. channel_axis: Index of axis for channels in the input tensor. fill_mode: Points outside the boundaries of the input are filled according to the given mode (one of `{'constant', 'nearest', 'reflect', 'wrap'}`). cval: Value used for points outside the boundaries of the input if `mode='constant'`. # Returns Shifted Numpy image tensor. ii(RRR R R R(RtwrgthrgRRRRRRRttxttyttranslation_matrixR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt random_shift7s c C@stjj| |}tjdtj| dgdtj|dgdddgg}|j||j|} } t|| | } t|| |||}|S(sPerforms a random spatial shear of a Numpy image tensor. # Arguments x: Input tensor. Must be 3D. intensity: Transformation intensity. row_axis: Index of axis for rows in the input tensor. col_axis: Index of axis for columns in the input tensor. channel_axis: Index of axis for channels in the input tensor. fill_mode: Points outside the boundaries of the input are filled according to the given mode (one of `{'constant', 'nearest', 'reflect', 'wrap'}`). cval: Value used for points outside the boundaries of the input if `mode='constant'`. # Returns Sheared Numpy image tensor. ii( RR R R R R RRR( Rt intensityRRRRRtsheart shear_matrixRRR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt random_shearWsc C@st|dkr$td|n|ddkrS|ddkrSd\}}n&tjj|d|dd\}}tj|ddgd|dgdddgg} |j||j|} } t| | | } t|| |||}|S(sPerforms a random spatial zoom of a Numpy image tensor. # Arguments x: Input tensor. Must be 3D. zoom_range: Tuple of floats; zoom range for width and height. row_axis: Index of axis for rows in the input tensor. col_axis: Index of axis for columns in the input tensor. channel_axis: Index of axis for channels in the input tensor. fill_mode: Points outside the boundaries of the input are filled according to the given mode (one of `{'constant', 'nearest', 'reflect', 'wrap'}`). cval: Value used for points outside the boundaries of the input if `mode='constant'`. # Returns Zoomed Numpy image tensor. # Raises ValueError: if `zoom_range` isn't a tuple. isBzoom_range should be a tuple or list of two floats. Received arg: ii(ii( tlent ValueErrorRR R R RRR( Rt zoom_rangeRRRRRtzxtzyt zoom_matrixRRR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt random_zoomus  & cC@stj||d}tj|tj|}}g|D]/}tj|tjj| |||^q;}tj|dd}tj|d|d}|S(Nitaxisi(RtrollaxistmintmaxtclipR R tstack(RR$Rtmin_xtmax_xt x_channeltchannel_images((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytrandom_channel_shifts9cC@st|dd}t|dd}tjdd|gdd|gdddgg}tjdd| gdd| gdddgg}tjtj|||}|S(Nig?ii(tfloatRR tdot(tmatrixRtyto_xto_yt offset_matrixt reset_matrixR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs 35c C@stj||d}|ddddf}|dddf}g|D]0}tjj|||ddd|d|^qN}tj|dd}tj|d|d}|S( sdApply the image transformation specified by a matrix. # Arguments x: 2D numpy array, single image. transform_matrix: Numpy array specifying the geometric transformation. channel_axis: Index of axis for channels in the input tensor. fill_mode: Points outside the boundaries of the input are filled according to the given mode (one of `{'constant', 'nearest', 'reflect', 'wrap'}`). cval: Value used for points outside the boundaries of the input if `mode='constant'`. # Returns The transformed version of the input. iNitordertmodeRR/i(RR0tndit interpolationtaffine_transformR4( RRRRRtfinal_affine_matrixt final_offsetR7R8((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs:cC@sJtj|j|d}|ddddf}|jd|}|S(Nii.(Rtasarraytswapaxes(RR/((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt flip_axisscC@stdkrtdntj|dtj}|jdkrZtd|j n|dkrutj }n|ddhkrtd|n|dkr|j dd d }n|r|t tj | d }tj |}|d kr||:}n|d 9}n|j d dkrAtj|jd d S|j d dkrtj|ddddd fjd dStd|j d dS(sConverts a 3D Numpy array to a PIL Image instance. # Arguments x: Input Numpy array. data_format: Image data format. scale: Whether to rescale image values to be within [0, 255]. # Returns A PIL Image instance. # Raises ImportError: if PIL is not available. ValueError: if invalid `x` or `data_format` is passed. sCCould not import PIL.Image. The use of `array_to_img` requires PIL.tdtypeisIExpected image array to have rank 3 (single image). Got array with shape:tchannels_firstt channels_lastsInvalid data_format:iiiituint8tRGBNtLsUnsupported channel number: (t pil_imagetNonet ImportErrorRRItKtfloatxtndimR)Rtimage_data_formatt transposeR2R1t fromarraytastype(Rt data_formattscaletx_max((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt array_to_imgs.      2cC@s|d krtj}n|ddhkr?td|ntj|dtj}t|jdkr|dkr|j ddd}qnt|jdkr |dkr|j d|jd|jdf}q|j |jd|jddf}ntd |j|S( sConverts a PIL Image instance to a Numpy array. # Arguments img: PIL Image instance. data_format: Image data format. # Returns A 3D Numpy array. # Raises ValueError: if invalid `img` or `data_format` is passed. RMRNsUnknown data_format: RLiiiisUnsupported image shape: N( RSRURXR)RRIRVR(RRYtreshape(timgR\R((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt img_to_array s   ))cC@stdkrtdntj|}|rT|jdkru|jd}qun!|jdkru|jd}n|r|d|df}|j|kr|j|}qn|S(syLoads an image into PIL format. # Arguments path: Path to image file grayscale: Boolean, whether to load the image as grayscale. target_size: Either `None` (default to original size) or tuple of ints `(img_height, img_width)`. # Returns A PIL Image instance. # Raises ImportError: if PIL is not available. sCCould not import PIL.Image. The use of `array_to_img` requires PIL.RQRPiiN(RRRSRTtopenRCtconverttsizetresize(tpatht grayscalet target_sizeRatwh_tuple((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytload_img.s sjpg|jpeg|bmp|pngcC@s_gtj|D]K\}}}|D]5}tjd|d|r#tjj||^q#qS(Ns ([\w]+\.(?:s))(tostwalktretmatchRgtjoin(t directorytexttroott_tfilestf((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt list_picturesNs&tImageDataGeneratorcB@seZdZeeeeeddddddddeeddddZddedddddZdd dd dedddded Zd Z d Z edddZ RS(sGenerate minibatches of image data with real-time data augmentation. # Arguments featurewise_center: set input mean to 0 over the dataset. samplewise_center: set each sample mean to 0. featurewise_std_normalization: divide inputs by std of the dataset. samplewise_std_normalization: divide each input by its std. zca_whitening: apply ZCA whitening. rotation_range: degrees (0 to 180). width_shift_range: fraction of total width. height_shift_range: fraction of total height. shear_range: shear intensity (shear angle in radians). zoom_range: amount of zoom. if scalar z, zoom will be randomly picked in the range [1-z, 1+z]. A sequence of two can be passed instead to select this range. channel_shift_range: shift range for each channels. fill_mode: points outside the boundaries are filled according to the given mode ('constant', 'nearest', 'reflect' or 'wrap'). Default is 'nearest'. cval: value used for points outside the boundaries when fill_mode is 'constant'. Default is 0. horizontal_flip: whether to randomly flip images horizontally. vertical_flip: whether to randomly flip images vertically. rescale: rescaling factor. If None or 0, no rescaling is applied, otherwise we multiply the data by the value provided (before applying any other transformation). preprocessing_function: function that will be implied on each input. The function will run before any other modification on it. The function should take one argument: one image (Numpy tensor with rank 3), and should output a Numpy tensor with the same shape. data_format: 'channels_first' or 'channels_last'. In 'channels_first' mode, the channels dimension (the depth) is at index 1, in 'channels_last' mode it is at index 3. It defaults to the `image_data_format` value found in your Keras config file at `~/.keras/keras.json`. If you never set it, then it will be "channels_last". gRcC@s|dkrtj}n||_||_||_||_||_||_||_ ||_ | |_ | |_ | |_ | |_| |_||_||_||_||_|ddhkrtd|n||_|dkr d|_d|_d|_n|dkr5d|_d|_d|_nd|_d|_d|_tj| ryd| d| g|_ n;t| dkr| d| dg|_ ntd| dS( NRNRMsdata_format should be "channels_last" (channel after row and column) or "channels_first" (channel before row and column). Received arg: iiiisMzoom_range should be a float or a tuple or list of two floats. Received arg: (RSRURXtfeaturewise_centertsamplewise_centertfeaturewise_std_normalizationtsamplewise_std_normalizationt zca_whiteningtrotation_rangetwidth_shift_rangetheight_shift_ranget shear_rangeR*tchannel_shift_rangeRRthorizontal_flipt vertical_fliptrescaletpreprocessing_functionR)R\RRRtmeantstdtprincipal_componentsRtisscalarR((tselfRyRzR{R|R}R~RRRR*RRRRRRRR\((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt__init__{sP                               i ttjpegc C@s=t|||d|d|d|d|jd|d|d|S(Nt batch_sizetshuffletseedR\t save_to_dirt save_prefixt save_format(tNumpyArrayIteratorR\( RRR=RRRRRR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytflows  itrgbt categoricalc C@sXt||d|d|d|d|d|jd|d|d|d | d | d | d |  S( NRit color_modetclassest class_modeR\RRRRRRt follow_links(tDirectoryIteratorR\( RRqRiRRRRRRRRRR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytflow_from_directorys    cC@s|jr|j|}n|jr4||j9}n|jd}|jrl|tj|d|dt8}n|jr|tj|d|dtd:}n|j r|jd k r||j8}qt j dn|j r|jd k r||jd:}qt j dn|jr|jd k rtj||j}tj||j}tj||jd|jd|jdf}qt j d n|S( sApply the normalization configuration to a batch of inputs. # Arguments x: batch of inputs to be normalized. # Returns The inputs, normalized. iR/tkeepdimsgHz>sThis ImageDataGenerator specifies `featurewise_center`, but it hasn'tbeen fit on any training data. Fit it first by calling `.fit(numpy_data)`.sThis ImageDataGenerator specifies `featurewise_std_normalization`, but it hasn'tbeen fit on any training data. Fit it first by calling `.fit(numpy_data)`.iisThis ImageDataGenerator specifies `zca_whitening`, but it hasn'tbeen fit on any training data. Fit it first by calling `.fit(numpy_data)`.N(RRRRzRRtTrueR|RRyRStwarningstwarnR{R}RR`ReR;R(RRtimg_channel_axistflatxtwhitex((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt standardizes0    " &   3c C@s"|jd}|jd}|jd}|jrZtjdtjj|j |j}nd}|jrtjj|j |j|j |}nd}|j rtjj|j |j |j |}nd}|j rtjj|j |j }nd}|j ddkr5|j ddkr5d\} } n,tjj|j d|j dd\} } d } |dkrtjtj|tj| dgtj|tj|dgdddgg} | } n|dks|dkrFtjdd|gdd|gdddgg} | d kr1| ntj| | } n|dkrtjdtj| dgdtj|dgdddgg}| d kr|ntj| |} n| dks| dkr1tj| ddgd| dgdddgg}| d kr|ntj| |} n| d k r|j ||j |}}t| ||} t|| |d|jd|j}n|jdkrt||j|}n|jrtjjdkrt||}qn|jrtjjdkrt||}qn|S( sRandomly augment a single image tensor. # Arguments x: 3D tensor, single image. # Returns A randomly transformed version of the input (same shape). iiiiRRg?(iiN(RRRR~RRR R RRRRR*RSR R R R;RRRRRR9RRKR(RRt img_row_axist img_col_axisRRR R!R%R+R,RRt shift_matrixR&R-RR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytrandom_transform sn    * * * &, %  ' ' '    icC@srtj|dtj}|jdkrFtdt|jn|j|jdddhkrtd|j dt|jdt|jd t|jd t|j|jd n|dk rtj j |ntj |}|rtjt||jd gt|jddtj}xXt|D]J}xAt|jd D],}|j||||||jd N(#RRIRURVRWR)tstrRRR\RSR RtcopytzerosttupletlistRRRyRRRR`R{RtepsilonR}R;tTRtsvdtdiagtsqrtR(RRtaugmenttroundsRtaxtrtitbroadcast_shapetflat_xtsigmatutsRt((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytfitbsBb @.  ' ' < N(ii( t__name__t __module__t__doc__tFalseRSRRRRRRR(((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRxTsB% 0    0 WtIteratorcB@sDeZdZdZdZdeddZdZdZ RS(s1Abstract base class for image data iterators. # Arguments n: Integer, total number of samples in the dataset to loop over. batch_size: Integer, size of a batch. shuffle: Boolean, whether to shuffle the data between epochs. seed: Random seeding for data shuffling. cC@s[||_||_||_d|_d|_tj|_|j|||||_ dS(Ni( tnRRt batch_indexttotal_batches_seent threadingtLocktlockt _flow_indextindex_generator(RRRRR((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs     cC@s d|_dS(Ni(R(R((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytresetsi cc@s|jx|dk r3tjj||jn|jdkrotj|}|rotjj|}qon|j||}|||kr|}|jd7_n||}d|_|jd7_||||!||fVq dS(Nii( RRSRR RRRtaranget permutation(RRRRRt index_arrayt current_indextcurrent_batch_size((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs"    cC@s|S(N((R((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt__iter__scO@s|j||S(N(tnext(Rtargstkwargs((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt__next__sN( RRRRRRRSRRR(((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs   RcB@s5eZdZdeddddddZdZRS(sIterator yielding data from a Numpy array. # Arguments x: Numpy array of input data. y: Numpy array of targets data. image_data_generator: Instance of `ImageDataGenerator` to use for random transformations and normalization. batch_size: Integer, size of a batch. shuffle: Boolean, whether to shuffle the data between epochs. seed: Random seed for data shuffling. data_format: String, one of `channels_first`, `channels_last`. save_to_dir: Optional directory where to save the pictures being yielded, in a viewable format. This is useful for visualizing the random transformations being applied, for debugging purposes. save_prefix: String prefix to use for saving sample images (if `save_to_dir` is set). save_format: Format to use for saving sample images (if `save_to_dir` is set). i RRc C@s|dk rUt|t|krUtdtj|jtj|jfn|dkrptj}ntj|dtj|_ |j j dkrtd|j jn|dkrdnd} |j j| dddhkrNtd|d t | d t | d t |j jd t |j j| d n|dk rotj||_ n d|_ ||_ ||_||_| |_| |_tt|j|jd|||dS(Ns_X (images tensor) and y (labels) should have the same length. Found: X.shape = %s, y.shape = %sRLisUInput data in `NumpyArrayIterator` should have rank 4. You passed an array with shapeRNiis=NumpyArrayIterator is set to use the data format convention "s" (channels on axis s3), i.e. expected either 1, 3 or 4 channels on axis s-. However, it was passed an array with shape s (s channels).i(RSR(R)RRIRRURXRVRRWRR=timage_data_generatorR\RRRtsuperRR( RRR=RRRRR\RRRt channels_axis((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRs,$+ \       c C@s|jt|j\}}}WdQXtjt|gt|jjddt j }xdt |D]V\}}|j|}|j j |jt j }|j j|}|||s(tsortedRlRm(tsubpath(R(s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyt_recursive_listst.s(Found %d images belonging to %d classes.RLtint32(i(i(i(i(*RSRURXRqRRRiR)RR\t image_shapeRRRRRtsamplesRRltlistdirRgtisdirRptappendR(t num_classtdicttzipRt class_indicesRtlowertendswithRtprintt filenamesRRtrelpathRRR(RRqRRiRRRRRRR\RRRRtwhite_list_formatstsubdirRRRsRtRuRtis_validt extensionRt absolute_path((Rs8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyRXs               !$        .c C@sY|jt|j\}}}WdQXtj|f|jdtj}|jdk}xt |D]\}}|j |}t t j j|j|d|d|j} t| d|j} |jj| } |jj| } | || %     &* (iiN(RRRRSRRRR(((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pyR5s!   Q(,Rt __future__RRtnumpyRRntscipyRt scipy.ndimagetndimageRDt six.movesRRlRRRRRUtPILRRRRTRSRR#R'R.R9RRRKRR_RbRRkRwtobjectRxRRR(((s8/tmp/pip-build-isqEY4/keras/keras/preprocessing/image.pytsL           '   2 " R5[