Ukubhekana nezinkinga ohlelweni noma ikhodi yokususa iphutha kuyinto evamile konjiniyela. Emkhakheni wohlelo lwePython, ukusebenza nemitapo yolwazi efana ne-Keras, i-API yokufunda ejulile, kuvame ukuletha izinselele. Enye yalezi I-Keras ImageDataGenerator ayisebenzi kahle. Leli thuluzi libaluleke kakhulu njengoba lisiza ekwandiseni idatha ngesikhathi sangempela. Nokho, uma yehluleka ukusebenza, iba yisithiyo ekuqhubekiseni phambili amaphrojekthi ethu okufunda ngomshini. Lesi sihloko sizosiza ukuxazulula inkinga futhi silungise inkinga ye-Keras ImageDataGenerator engasebenzi kahle.
I-Keras ithandwa kakhulu ngabathuthukisi ngenxa yokuthi ilula ukuyisebenzisa futhi ilula ukuyisebenzisa, okwenza kube lula ukudala amamodeli okufunda okujulile ngaphandle kokudinga ukuqonda ngokuphelele izakhiwo eziyinkimbinkimbi eziyisisekelo. Kodwa, noma ngabe kulula ukuyisebenzisa, umuntu angavame ukubhekana nezinkinga ngayo, okuholela ezinseleleni zokwakha amamodeli anembile ngempumelelo.
Ibhekana Nenkinga Ye-ImageDataGenerator
Ngaphambi kokungena esixazululweni , kubalulekile ukuqonda i-ImageDataGenerator kanye nendima yayo. Isetshenziswa kakhulu ekukhuliseni idatha okuyindlela ephumelelayo yokwandisa usayizi wesethi yedatha nokugwema ukufaka ngokweqile kumamodeli wenethiwekhi ye-neural. Uma lokhu kungasebenzi, ukunemba kwemodeli yakho kungase kungabi phezulu njengoba ubulindele.
Ukuze uxazulule le nkinga, kufanele uqale uqinisekise ukuthi izindlela zakho zefayela zilungile yini. Izindlela zefayela ezingalungile zingenye yezizathu ezivame kakhulu ze-ImageDataGenerator ingasebenzi kahle.
from keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator(
rotation_range=40,
width_shift_range=0.2,
height_shift_range=0.2,
rescale=1./255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest')
# Should print 'Found x images belonging to y classes.'
train_generator = datagen.flow_from_directory(
'path_to_train_directory', # Double check this path
target_size=(150, 150),
batch_size=32,
class_mode='binary')
Ikhodi Incazelo Yesinyathelo Nesinyathelo
Isinyathelo sokuqala ukungenisa umtapo wolwazi odingekayo, okuyikilasi le-ImageDataGenerator ukusuka ku-keras.preprocessing.image. I-ImageDataGenerator oyiqalisayo inesethi yamapharamitha ambalwa, amele izindlela idatha yesithombe engashintshwa ngayo ngokungahleliwe ukuze ikhulise.
Into ye-train_generator kamuva idalwa ngokunxenxa umsebenzi we-flow_from_directory entweni yedathagen namapharamitha athile: umzila wohla lwemibhalo oqukethe idatha yakho yokuqeqeshwa, usayizi oqondiwe onquma ubukhulu lapho zonke izithombe ezitholiwe zizoshintshwa khona usayizi, usayizi weqoqo, kanye neklasi. imodi.
Ingabe singaphakamisa ukuthi uphinde uhlole indlela yakho eya kuhla lwemibhalo lokuqeqesha? Kwesinye isikhathi, iphutha elincane lingaba yimbangela yokuthi i-generator ingasebenzi.
Ukuvakashela kabusha ama-Keras kanye ne-ImageDataGenerator
Onjiniyela abasebenza kakhulu ekuklameni amamodeli okufunda ajulile bangathola i-Keras umtapo wolwazi owusizo ngempela. Amandla ayo abalulekile alele ekululameni kwayo nokuthi ihlanganisa kanjani ubunkimbinkimbi bokwakha amamodeli okufunda ajulile kuyilapho inikeza isikhombimsebenzisi esibonakalayo kubahleli bohlelo.
I-ImageDataGenerator ikhanya ikakhulukazi uma unedatha elinganiselwe. Ukwengeza ngokuzenzakalelayo idathasethi ngokusebenzisa ukuguqulwa kwesithombe okungahleliwe kusiza ukwenza imodeli ingaguquguquki kulokhu kuguqulwa futhi kuvumela ukwenziwa okuvamile okungcono.
Ukusetshenziswa okusabalele kwe-Keras kanye nomsebenzi wayo we-ImageDataGenerator, indima yawo ebalulekile ekuthuthukisweni kwemodeli yokufunda ejulile, kanye nokuxazululwa okulula kobunzima obuyizungezile, konke kuwubufakazi benzuzo yalo mtapo wolwazi, emhlabeni we-AI.
