Preface
Previous articleTensorFlow case study: classifying clothing images In this article, we followed the official documentation and learned how to preprocess data and build Models, training models, etc. But training a model is very difficult for amateur players like me. So why don’t we stand on the shoulders of giants and use mature models that have already been trained? This article briefly introduces how to use mature models
It is recommended to watch in order, this is a small series suitable for beginners like me to get started
Configuration environment:Tensorflow installation in windows environment
Image classification case study:TensorFlow case study: Classifying clothing images
Use the official model and fine-tune:TensorFlow learning: use the official model for image classification and use your own data to fine-tune the model
Convert the model and use it on the front end:TensorFlow learning: How to use the Keras model on the web front end
Use mature models
use
Download model
The official provides us with a model download website suitable forTensorFlow: TensorFlow Hub (requires scientific Internet access)
The model I chose here is:mobilenet_v2, and the model download address ishttps://tfhub .dev/google/imagenet/mobilenet_v2_140_224/classification/5
Model introduction
MobileNetV2 is a lightweight convolutional neural network architecture developed by Google, specifically designed for image classification and object detection tasks on mobile devices and embedded devices. MobileNetV2 is the second version of the MobileNet series, which achieves a better balance between accuracy and performance than the previous version.
MobileNetV2Achieved comparable accuracy to larger and more complex models onImageNetimage classification tasks, but with significantly reduced model size and computational effort. This makesMobileNetV2ideal for real-time image classification and object detection on resource-constrained devices.
In TensorFlow, you can use pre-trainedMobileNetV2 models, or you can fine-tune and train according to your own needs. TensorFlow provides corresponding APIs and tools to make use and training more convenient. MobileNetV2
Use
to unzip the downloaded model and import it into the project, complete code
# 导入tensorflow 和科学计算库
import tensorflow as tf
import numpy as np
# tensorflow-hub是一个TensorFlow库的扩展，它提供了一个简单的接口，用于重用已经训练好的机器学习模型的部分
import tensorflow_hub as hub
# 字体属性
from matplotlib.font_manager import FontProperties
# matplotlib是用于绘制图表和可视化数据的库
import matplotlib.pylab as plt
# 用于加载json文件
import json
# 导入模型
# 不能直接加载模型文件，需要加载器目录
# 加载mobilenet_v2模型，这里要加载文件夹不要直接加载pb文件
# 模型如何加载要看文档，原来使用tf.keras.models.load_model加载一直失败
model = tf.keras.Sequential([
hub.KerasLayer('imagenet_mobilenet_v2_140_224_classification_5')
])
print("模型信息:",model)
# 预处理输入数据
# 1、mobilenet需要的图片尺寸是 224 * 224
image = tf.keras.preprocessing.image.load_img('pics/dog.png',target_size=(224,224))
# 2、将图片转为数组，既是只有一张图片
image = tf.keras.preprocessing.image.img_to_array(image)
# 3、扩展数组维度，使其符合模型的输入
image = np.expand_dims(image, axis=0)
# 4、使用mobilenet_v2提供的预处理函数对图像处理，包括图像归一化、颜色通道顺序调整、像素值标准化等操作
image = tf.keras.applications.mobilenet_v2.preprocess_input(image)
# 预测
predictions = model.predict(image)
# 获取最高概率对应的类别索引
predicted_index = np.argmax(predictions)
# 概率值
confidence = np.max(predictions)
print("索引和概率值是：",predicted_index,confidence)
# 加载映射文件
with open('imagenet-classes.json','r') as f:
labels_dict = json.load(f)
# 类别的索引是字符串，这里要简单处理一下，这里-1是因为官方提供的多了一个0（背景），我找到的标签没有这个，因此要-1
class_name = labels_dict[str(predicted_index-1)]
# 可视化显示
font = FontProperties()
font.set_family('Microsoft YaHei')
plt.figure() # 创建图像窗口
plt.xticks([])
plt.yticks([])
plt.grid(False) # 取消网格线
plt.imshow(image[0]) # 显示图片
plt.xlabel(class_name,fontproperties=font)
plt.show() # 显示图形窗口
important point
Remove some unnecessary code. In fact, there is really very little code. But you can use these codes to classify images. Although there is not much code, many problems have been encountered.
Loading model
I encountered a problem while loading the model. This was a slap in the face.
When loading the model first, you need to load the entire folder, not a file within the file
model = tf.keras.Sequential([
hub.KerasLayer('imagenet_mobilenet_v2_140_224_classification_5')
])
The folder is as follows:
The function used to load the model. The function for loading the model found by Baidu at the beginning was:tf.keras.models.load_model. But it does not apply here, causing the model to be loaded unsuccessfully. Later, I looked at the official documentation and found that it needs to be loaded like this
Here we want to downloadtensorflow_hub, in the loading command hub is its abbreviation. tensorflow-hub is an extension of the TensorFlow library that provides a simple interface for reusing parts of already trained machine learning models. Different models have different loading methods. When using the models provided by the TensorFlow Hub website, please pay attention to their instructions.
pip install tensorflow_hub
Preprocess input data
Loading the model is not enough, you also need to process the data. Make your image conform to the input format of the model, which is the following code
# 预处理输入数据
# 1、mobilenet需要的图片尺寸是 224 * 224
image = tf.keras.preprocessing.image.load_img('pics/dog.png',target_size=(224,224))
# 2、将图片转为数组，既是只有一张图片
image = tf.keras.preprocessing.image.img_to_array(image)
# 3、扩展数组维度，使其符合模型的输入
image = np.expand_dims(image, axis=0)
# 4、使用mobilenet_v2提供的预处理函数对图像处理，包括图像归一化、颜色通道顺序调整、像素值标准化等操作
image = tf.keras.applications.mobilenet_v2.preprocess_input(image)
Mapping file
After the prediction is completed, you need to know what exactly is this image? The model is trained based on the Imagenet (ILSVRC-2012-CLS) data set, so you need to obtain the corresponding mapping file. I found it here from the Internet, but there is a slight difference
The official mapping file adds a 0 at the beginning, but the mapping file I found does not. If the serial number is modified, there will be too much data, so the serial number is manually reduced by -1 when obtaining the category.class_name = labels_dict[str(predicted_index-1)]
Mapping file:
{
"-1":["background"],"0":["n01440764","tench"],"1":["n01443537","goldfish"],"2":["n01484850","great white shark"],"3":["n01491361","tiger shark"],"4":["n01494475","hammerhead"],"5":["n01496331","electric ray"],"6":["n01498041","stingray"],"7":["n01514668","cock"],"8":["n01514859","hen"],"9":["n01518878","ostrich"],"10":["n01530575","brambling"],"11":["n01531178","goldfinch"],"12":["n01532829","house finch"],"13":["n01534433","junco"],"14":["n01537544","indigo bunting"],"15":["n01558993","robin"],"16":["n01560419","bulbul"],"17":["n01580077","jay"],"18":["n01582220","magpie"],"19":["n01592084","chickadee"],"20":["n01601694","water ouzel"],"21":["n01608432","kite"],"22":["n01614925","bald eagle"],"23":["n01616318","vulture"],"24":["n01622779","great grey owl"],"25":["n01629819","European fire salamander"],"26":["n01630670","common newt"],"27":["n01631663","eft"],"28":["n01632458","spotted salamander"],"29":["n01632777","axolotl"],"30":["n01641577","bullfrog"],"31":["n01644373","tree frog"],"32":["n01644900","tailed frog"],"33":["n01664065","loggerhead"],"34":["n01665541","leatherback turtle"],"35":["n01667114","mud turtle"],"36":["n01667778","terrapin"],"37":["n01669191","box turtle"],"38":["n01675722","banded gecko"],"39":["n01677366","common iguana"],"40":["n01682714","American chameleon"],"41":["n01685808","whiptail"],"42":["n01687978","agama"],"43":["n01688243","frilled lizard"],"44":["n01689811","alligator lizard"],"45":["n01692333","Gila monster"],"46":["n01693334","green lizard"],"47":["n01694178","African chameleon"],"48":["n01695060","Komodo dragon"],"49":["n01697457","African crocodile"],"50":["n01698640","American alligator"],"51":["n01704323","triceratops"],"52":["n01728572","thunder snake"],"53":["n01728920","ringneck snake"],"54":["n01729322","hognose snake"],"55":["n01729977","green snake"],"56":["n01734418","king snake"],"57":["n01735189","garter snake"],"58":["n01737021","water snake"],"59":["n01739381","vine snake"],"60":["n01740131","night snake"],"61":["n01742172","boa constrictor"],"62":["n01744401","rock python"],"63":["n01748264","Indian cobra"],"64":["n01749939","green mamba"],"65":["n01751748","sea snake"],"66":["n01753488","horned viper"],"67":["n01755581","diamondback"],"68":["n01756291","sidewinder"],"69":["n01768244","trilobite"],"70":["n01770081","harvestman"],"71":["n01770393","scorpion"],"72":["n01773157","black and gold garden spider"],"73":["n01773549","barn spider"],"74":["n01773797","garden spider"],"75":["n01774384","black widow"],"76":["n01774750","tarantula"],"77":["n01775062","wolf spider"],"78":["n01776313","tick"],"79":["n01784675","centipede"],"80":["n01795545","black grouse"],"81":["n01796340","ptarmigan"],"82":["n01797886","ruffed grouse"],"83":["n01798484","prairie chicken"],"84":["n01806143","peacock"],"85":["n01806567","quail"],"86":["n01807496","partridge"],"87":["n01817953","African grey"],"88":["n01818515","macaw"],"89":["n01819313","sulphur-crested cockatoo"],"90":["n01820546","lorikeet"],"91":["n01824575","coucal"],"92":["n01828970","bee eater"],"93":["n01829413","hornbill"],"94":["n01833805","hummingbird"],"95":["n01843065","jacamar"],"96":["n01843383","toucan"],"97":["n01847000","drake"],"98":["n01855032","red-breasted merganser"],"99":["n01855672","goose"],"100":["n01860187","black swan"],
"101":["n01871265","tusker"],"102":["n01872401","echidna"],"103":["n01873310","platypus"],"104":["n01877812","wallaby"],"105":["n01882714","koala"],"106":["n01883070","wombat"],"107":["n01910747","jellyfish"],"108":["n01914609","sea anemone"],"109":["n01917289","brain coral"],"110":["n01924916","flatworm"],"111":["n01930112","nematode"],"112":["n01943899","conch"],"113":["n01944390","snail"],"114":["n01945685","slug"],"115":["n01950731","sea slug"],"116":["n01955084","chiton"],"117":["n01968897","chambered nautilus"],"118":["n01978287","Dungeness crab"],"119":["n01978455","rock crab"],"120":["n01980166","fiddler crab"],"121":["n01981276","king crab"],"122":["n01983481","American lobster"],"123":["n01984695","spiny lobster"],"124":["n01985128","crayfish"],"125":["n01986214","hermit crab"],"126":["n01990800","isopod"],"127":["n02002556","white stork"],"128":["n02002724","black stork"],"129":["n02006656","spoonbill"],"130":["n02007558","flamingo"],"131":["n02009229","little blue heron"],"132":["n02009912","American egret"],"133":["n02011460","bittern"],"134":["n02012849","crane"],"135":["n02013706","limpkin"],"136":["n02017213","European gallinule"],"137":["n02018207","American coot"],"138":["n02018795","bustard"],"139":["n02025239","ruddy turnstone"],"140":["n02027492","red-backed sandpiper"],"141":["n02028035","redshank"],"142":["n02033041","dowitcher"],"143":["n02037110","oystercatcher"],"144":["n02051845","pelican"],"145":["n02056570","king penguin"],"146":["n02058221","albatross"],"147":["n02066245","grey whale"],"148":["n02071294","killer whale"],"149":["n02074367","dugong"],"150":["n02077923","sea lion"],"151":["n02085620","Chihuahua"],"152":["n02085782","Japanese spaniel"],"153":["n02085936","Maltese dog"],"154":["n02086079","Pekinese"],"155":["n02086240","Shih-Tzu"],"156":["n02086646","Blenheim spaniel"],"157":["n02086910","papillon"],"158":["n02087046","toy terrier"],"159":["n02087394","Rhodesian ridgeback"],"160":["n02088094","Afghan hound"],"161":["n02088238","basset"],"162":["n02088364","beagle"],"163":["n02088466","bloodhound"],"164":["n02088632","bluetick"],"165":["n02089078","black-and-tan coonhound"],"166":["n02089867","Walker hound"],"167":["n02089973","English foxhound"],"168":["n02090379","redbone"],"169":["n02090622","borzoi"],"170":["n02090721","Irish wolfhound"],"171":["n02091032","Italian greyhound"],"172":["n02091134","whippet"],"173":["n02091244","Ibizan hound"],"174":["n02091467","Norwegian elkhound"],"175":["n02091635","otterhound"],"176":["n02091831","Saluki"],"177":["n02092002","Scottish deerhound"],"178":["n02092339","Weimaraner"],"179":["n02093256","Staffordshire bullterrier"],"180":["n02093428","American Staffordshire terrier"],"181":["n02093647","Bedlington terrier"],"182":["n02093754","Border terrier"],"183":["n02093859","Kerry blue terrier"],"184":["n02093991","Irish terrier"],"185":["n02094114","Norfolk terrier"],"186":["n02094258","Norwich terrier"],"187":["n02094433","Yorkshire terrier"],"188":["n02095314","wire-haired fox terrier"],"189":["n02095570","Lakeland terrier"],"190":["n02095889","Sealyham terrier"],"191":["n02096051","Airedale"],"192":["n02096177","cairn"],"193":["n02096294","Australian terrier"],"194":["n02096437","Dandie Dinmont"],"195":["n02096585","Boston bull"],"196":["n02097047","miniature schnauzer"],
"197":["n02097130","giant schnauzer"],"198":["n02097209","standard schnauzer"],"199":["n02097298","Scotch terrier"],"200":["n02097474","Tibetan terrier"],"201":["n02097658","silky terrier"],"202":["n02098105","soft-coated wheaten terrier"],"203":["n02098286","West Highland white terrier"],"204":["n02098413","Lhasa"],"205":["n02099267","flat-coated retriever"],"206":["n02099429","curly-coated retriever"],"207":["n02099601","golden retriever"],"208":["n02099712","Labrador retriever"],"209":["n02099849","Chesapeake Bay retriever"],"210":["n02100236","German short-haired pointer"],"211":["n02100583","vizsla"],"212":["n02100735","English setter"],"213":["n02100877","Irish setter"],"214":["n02101006","Gordon setter"],"215":["n02101388","Brittany spaniel"],"216":["n02101556","clumber"],"217":["n02102040","English springer"],"218":["n02102177","Welsh springer spaniel"],"219":["n02102318","cocker spaniel"],"220":["n02102480","Sussex spaniel"],"221":["n02102973","Irish water spaniel"],"222":["n02104029","kuvasz"],"223":["n02104365","schipperke"],"224":["n02105056","groenendael"],"225":["n02105162","malinois"],"226":["n02105251","briard"],"227":["n02105412","kelpie"],"228":["n02105505","komondor"],"229":["n02105641","Old English sheepdog"],"230":["n02105855","Shetland sheepdog"],"231":["n02106030","collie"],"232":["n02106166","Border collie"],"233":["n02106382","Bouvier des Flandres"],"234":["n02106550","Rottweiler"],"235":["n02106662","German shepherd"],"236":["n02107142","Doberman"],"237":["n02107312","miniature pinscher"],"238":["n02107574","Greater Swiss Mountain dog"],"239":["n02107683","Bernese mountain dog"],"240":["n02107908","Appenzeller"],"241":["n02108000","EntleBucher"],"242":["n02108089","boxer"],"243":["n02108422","bull mastiff"],"244":["n02108551","Tibetan mastiff"],"245":["n02108915","French bulldog"],"246":["n02109047","Great Dane"],"247":["n02109525","Saint Bernard"],"248":["n02109961","Eskimo dog"],"249":["n02110063","malamute"],"250":["n02110185","Siberian husky"],"251":["n02110341","dalmatian"],"252":["n02110627","affenpinscher"],"253":["n02110806","basenji"],"254":["n02110958","pug"],"255":["n02111129","Leonberg"],"256":["n02111277","Newfoundland"],"257":["n02111500","Great Pyrenees"],"258":["n02111889","Samoyed"],"259":["n02112018","Pomeranian"],"260":["n02112137","chow"],"261":["n02112350","keeshond"],"262":["n02112706","Brabancon griffon"],"263":["n02113023","Pembroke"],"264":["n02113186","Cardigan"],"265":["n02113624","toy poodle"],"266":["n02113712","miniature poodle"],"267":["n02113799","standard poodle"],"268":["n02113978","Mexican hairless"],"269":["n02114367","timber wolf"],"270":["n02114548","white wolf"],"271":["n02114712","red wolf"],"272":["n02114855","coyote"],"273":["n02115641","dingo"],"274":["n02115913","dhole"],"275":["n02116738","African hunting dog"],"276":["n02117135","hyena"],"277":["n02119022","red fox"],"278":["n02119789","kit fox"],"279":["n02120079","Arctic fox"],"280":["n02120505","grey fox"],"281":["n02123045","tabby"],"282":["n02123159","tiger cat"],"283":["n02123394","Persian cat"],"284":["n02123597","Siamese cat"],"285":["n02124075","Egyptian cat"],"286":["n02125311","cougar"],"287":["n02127052","lynx"],"288":["n02128385","leopard"],"289":["n02128757","snow leopard"],"290":["n02128925","jaguar"],
"291":["n02129165","lion"],"292":["n02129604","tiger"],"293":["n02130308","cheetah"],"294":["n02132136","brown bear"],"295":["n02133161","American black bear"],"296":["n02134084","ice bear"],"297":["n02134418","sloth bear"],"298":["n02137549","mongoose"],"299":["n02138441","meerkat"],"300":["n02165105","tiger beetle"],"301":["n02165456","ladybug"],"302":["n02167151","ground beetle"],"303":["n02168699","long-horned beetle"],"304":["n02169497","leaf beetle"],"305":["n02172182","dung beetle"],"306":["n02174001","rhinoceros beetle"],"307":["n02177972","weevil"],"308":["n02190166","fly"],"309":["n02206856","bee"],"310":["n02219486","ant"],"311":["n02226429","grasshopper"],"312":["n02229544","cricket"],"313":["n02231487","walking stick"],"314":["n02233338","cockroach"],"315":["n02236044","mantis"],"316":["n02256656","cicada"],"317":["n02259212","leafhopper"],"318":["n02264363","lacewing"],"319":["n02268443","dragonfly"],"320":["n02268853","damselfly"],"321":["n02276258","admiral"],"322":["n02277742","ringlet"],"323":["n02279972","monarch"],"324":["n02280649","cabbage butterfly"],"325":["n02281406","sulphur butterfly"],"326":["n02281787","lycaenid"],"327":["n02317335","starfish"],"328":["n02319095","sea urchin"],"329":["n02321529","sea cucumber"],"330":["n02325366","wood rabbit"],"331":["n02326432","hare"],"332":["n02328150","Angora"],"333":["n02342885","hamster"],"334":["n02346627","porcupine"],"335":["n02356798","fox squirrel"],"336":["n02361337","marmot"],"337":["n02363005","beaver"],"338":["n02364673","guinea pig"],"339":["n02389026","sorrel"],"340":["n02391049","zebra"],"341":["n02395406","hog"],"342":["n02396427","wild boar"],"343":["n02397096","warthog"],"344":["n02398521","hippopotamus"],"345":["n02403003","ox"],"346":["n02408429","water buffalo"],"347":["n02410509","bison"],"348":["n02412080","ram"],"349":["n02415577","bighorn"],"350":["n02417914","ibex"],"351":["n02422106","hartebeest"],"352":["n02422699","impala"],"353":["n02423022","gazelle"],"354":["n02437312","Arabian camel"],"355":["n02437616","llama"],"356":["n02441942","weasel"],"357":["n02442845","mink"],"358":["n02443114","polecat"],"359":["n02443484","black-footed ferret"],"360":["n02444819","otter"],"361":["n02445715","skunk"],"362":["n02447366","badger"],"363":["n02454379","armadillo"],"364":["n02457408","three-toed sloth"],"365":["n02480495","orangutan"],"366":["n02480855","gorilla"],"367":["n02481823","chimpanzee"],"368":["n02483362","gibbon"],"369":["n02483708","siamang"],"370":["n02484975","guenon"],"371":["n02486261","patas"],"372":["n02486410","baboon"],"373":["n02487347","macaque"],"374":["n02488291","langur"],"375":["n02488702","colobus"],"376":["n02489166","proboscis monkey"],"377":["n02490219","marmoset"],"378":["n02492035","capuchin"],"379":["n02492660","howler monkey"],"380":["n02493509","titi"],"381":["n02493793","spider monkey"],"382":["n02494079","squirrel monkey"],"383":["n02497673","Madagascar cat"],"384":["n02500267","indri"],"385":["n02504013","Indian elephant"],"386":["n02504458","African elephant"],"387":["n02509815","lesser panda"],"388":["n02510455","giant panda"],"389":["n02514041","barracouta"],"390":["n02526121","eel"],"391":["n02536864","coho"],"392":["n02606052","rock beauty"],"393":["n02607072","anemone fish"],"394":["n02640242","sturgeon"],
"395":["n02641379","gar"],"396":["n02643566","lionfish"],"397":["n02655020","puffer"],"398":["n02666196","abacus"],"399":["n02667093","abaya"],"400":["n02669723","academic gown"],"401":["n02672831","accordion"],"402":["n02676566","acoustic guitar"],"403":["n02687172","aircraft carrier"],"404":["n02690373","airliner"],"405":["n02692877","airship"],"406":["n02699494","altar"],"407":["n02701002","ambulance"],"408":["n02704792","amphibian"],"409":["n02708093","analog clock"],"410":["n02727426","apiary"],"411":["n02730930","apron"],"412":["n02747177","ashcan"],"413":["n02749479","assault rifle"],"414":["n02769748","backpack"],"415":["n02776631","bakery"],"416":["n02777292","balance beam"],"417":["n02782093","balloon"],"418":["n02783161","ballpoint"],"419":["n02786058","Band Aid"],"420":["n02787622","banjo"],"421":["n02788148","bannister"],"422":["n02790996","barbell"],"423":["n02791124","barber chair"],"424":["n02791270","barbershop"],"425":["n02793495","barn"],"426":["n02794156","barometer"],"427":["n02795169","barrel"],"428":["n02797295","barrow"],"429":["n02799071","baseball"],"430":["n02802426","basketball"],"431":["n02804414","bassinet"],"432":["n02804610","bassoon"],"433":["n02807133","bathing cap"],"434":["n02808304","bath towel"],"435":["n02808440","bathtub"],"436":["n02814533","beach wagon"],"437":["n02814860","beacon"],"438":["n02815834","beaker"],"439":["n02817516","bearskin"],"440":["n02823428","beer bottle"],"441":["n02823750","beer glass"],"442":["n02825657","bell cote"],"443":["n02834397","bib"],"444":["n02835271","bicycle-built-for-two"],"445":["n02837789","bikini"],"446":["n02840245","binder"],"447":["n02841315","binoculars"],"448":["n02843684","birdhouse"],"449":["n02859443","boathouse"],"450":["n02860847","bobsled"],"451":["n02865351","bolo tie"],"452":["n02869837","bonnet"],"453":["n02870880","bookcase"],"454":["n02871525","bookshop"],"455":["n02877765","bottlecap"],"456":["n02879718","bow"],"457":["n02883205","bow tie"],"458":["n02892201","brass"],"459":["n02892767","brassiere"],"460":["n02894605","breakwater"],"461":["n02895154","breastplate"],"462":["n02906734","broom"],"463":["n02909870","bucket"],"464":["n02910353","buckle"],"465":["n02916936","bulletproof vest"],"466":["n02917067","bullet train"],"467":["n02927161","butcher shop"],"468":["n02930766","cab"],"469":["n02939185","caldron"],"470":["n02948072","candle"],"471":["n02950826","cannon"],"472":["n02951358","canoe"],"473":["n02951585","can opener"],"474":["n02963159","cardigan"],"475":["n02965783","car mirror"],"476":["n02966193","carousel"],"477":["n02966687","carpenter's kit"],"478":["n02971356","carton"],"479":["n02974003","car wheel"],"480":["n02977058","cash machine"],"481":["n02978881","cassette"],"482":["n02979186","cassette player"],"483":["n02980441","castle"],"484":["n02981792","catamaran"],"485":["n02988304","CD player"],"486":["n02992211","cello"],"487":["n02992529","cellular telephone"],"488":["n02999410","chain"],"489":["n03000134","chainlink fence"],"490":["n03000247","chain mail"],"491":["n03000684","chain saw"],"492":["n03014705","chest"],"493":["n03016953","chiffonier"],"494":["n03017168","chime"],"495":["n03018349","china cabinet"],"496":["n03026506","Christmas stocking"],"497":["n03028079","church"],"498":["n03032252","cinema"],"499":["n03041632","cleaver"],
"500":["n03042490","cliff dwelling"],"501":["n03045698","cloak"],"502":["n03047690","clog"],"503":["n03062245","cocktail shaker"],"504":["n03063599","coffee mug"],"505":["n03063689","coffeepot"],"506":["n03065424","coil"],"507":["n03075370","combination lock"],"508":["n03085013","computer keyboard"],"509":["n03089624","confectionery"],"510":["n03095699","container ship"],"511":["n03100240","convertible"],"512":["n03109150","corkscrew"],"513":["n03110669","cornet"],"514":["n03124043","cowboy boot"],"515":["n03124170","cowboy hat"],"516":["n03125729","cradle"],"517":["n03126707","crane"],"518":["n03127747","crash helmet"],"519":["n03127925","crate"],"520":["n03131574","crib"],"521":["n03133878","Crock Pot"],"522":["n03134739","croquet ball"],"523":["n03141823","crutch"],"524":["n03146219","cuirass"],"525":["n03160309","dam"],"526":["n03179701","desk"],"527":["n03180011","desktop computer"],"528":["n03187595","dial telephone"],"529":["n03188531","diaper"],"530":["n03196217","digital clock"],"531":["n03197337","digital watch"],"532":["n03201208","dining table"],"533":["n03207743","dishrag"],"534":["n03207941","dishwasher"],"535":["n03208938","disk brake"],"536":["n03216828","dock"],"537":["n03218198","dogsled"],"538":["n03220513","dome"],"539":["n03223299","doormat"],"540":["n03240683","drilling platform"],"541":["n03249569","drum"],"542":["n03250847","drumstick"],"543":["n03255030","dumbbell"],"544":["n03259280","Dutch oven"],"545":["n03271574","electric fan"],"546":["n03272010","electric guitar"],"547":["n03272562","electric locomotive"],"548":["n03290653","entertainment center"],"549":["n03291819","envelope"],"550":["n03297495","espresso maker"],"551":["n03314780","face powder"],"552":["n03325584","feather boa"],"553":["n03337140","file"],"554":["n03344393","fireboat"],"555":["n03345487","fire engine"],"556":["n03347037","fire screen"],"557":["n03355925","flagpole"],"558":["n03372029","flute"],"559":["n03376595","folding chair"],"560":["n03379051","football helmet"],"561":["n03384352","forklift"],"562":["n03388043","fountain"],"563":["n03388183","fountain pen"],"564":["n03388549","four-poster"],"565":["n03393912","freight car"],"566":["n03394916","French horn"],"567":["n03400231","frying pan"],"568":["n03404251","fur coat"],"569":["n03417042","garbage truck"],"570":["n03424325","gasmask"],"571":["n03425413","gas pump"],"572":["n03443371","goblet"],"573":["n03444034","go-kart"],"574":["n03445777","golf ball"],"575":["n03445924","golfcart"],"576":["n03447447","gondola"],"577":["n03447721","gong"],"578":["n03450230","gown"],"579":["n03452741","grand piano"],"580":["n03457902","greenhouse"],"581":["n03459775","grille"],"582":["n03461385","grocery store"],"583":["n03467068","guillotine"],"584":["n03476684","hair slide"],"585":["n03476991","hair spray"],"586":["n03478589","half track"],"587":["n03481172","hammer"],"588":["n03482405","hamper"],"589":["n03483316","hand blower"],"590":["n03485407","hand-held computer"],"591":["n03485794","handkerchief"],"592":["n03492542","hard disc"],"593":["n03494278","harmonica"],"594":["n03495258","harp"],"595":["n03496892","harvester"],"596":["n03498962","hatchet"],"597":["n03527444","holster"],"598":["n03529860","home theater"],"599":["n03530642","honeycomb"],"600":["n03532672","hook"],"601":["n03534580","hoopskirt"],
"602":["n03535780","horizontal bar"],"603":["n03538406","horse cart"],"604":["n03544143","hourglass"],"605":["n03584254","iPod"],"606":["n03584829","iron"],"607":["n03590841","jack-o'-lantern"],"608":["n03594734","jean"],"609":["n03594945","jeep"],"610":["n03595614","jersey"],"611":["n03598930","jigsaw puzzle"],"612":["n03599486","jinrikisha"],"613":["n03602883","joystick"],"614":["n03617480","kimono"],"615":["n03623198","knee pad"],"616":["n03627232","knot"],"617":["n03630383","lab coat"],"618":["n03633091","ladle"],"619":["n03637318","lampshade"],"620":["n03642806","laptop"],"621":["n03649909","lawn mower"],"622":["n03657121","lens cap"],"623":["n03658185","letter opener"],"624":["n03661043","library"],"625":["n03662601","lifeboat"],"626":["n03666591","lighter"],"627":["n03670208","limousine"],"628":["n03673027","liner"],"629":["n03676483","lipstick"],"630":["n03680355","Loafer"],"631":["n03690938","lotion"],"632":["n03691459","loudspeaker"],"633":["n03692522","loupe"],"634":["n03697007","lumbermill"],"635":["n03706229","magnetic compass"],"636":["n03709823","mailbag"],"637":["n03710193","mailbox"],"638":["n03710637","maillot"],"639":["n03710721","maillot"],"640":["n03717622","manhole cover"],"641":["n03720891","maraca"],"642":["n03721384","marimba"],"643":["n03724870","mask"],"644":["n03729826","matchstick"],"645":["n03733131","maypole"],"646":["n03733281","maze"],"647":["n03733805","measuring cup"],"648":["n03742115","medicine chest"],"649":["n03743016","megalith"],"650":["n03759954","microphone"],"651":["n03761084","microwave"],"652":["n03763968","military uniform"],"653":["n03764736","milk can"],"654":["n03769881","minibus"],"655":["n03770439","miniskirt"],"656":["n03770679","minivan"],"657":["n03773504","missile"],"658":["n03775071","mitten"],"659":["n03775546","mixing bowl"],"660":["n03776460","mobile home"],"661":["n03777568","Model T"],"662":["n03777754","modem"],"663":["n03781244","monastery"],"664":["n03782006","monitor"],"665":["n03785016","moped"],"666":["n03786901","mortar"],"667":["n03787032","mortarboard"],"668":["n03788195","mosque"],"669":["n03788365","mosquito net"],"670":["n03791053","motor scooter"],"671":["n03792782","mountain bike"],"672":["n03792972","mountain tent"],"673":["n03793489","mouse"],"674":["n03794056","mousetrap"],"675":["n03796401","moving van"],"676":["n03803284","muzzle"],"677":["n03804744","nail"],"678":["n03814639","neck brace"],"679":["n03814906","necklace"],"680":["n03825788","nipple"],"681":["n03832673","notebook"],"682":["n03837869","obelisk"],"683":["n03838899","oboe"],"684":["n03840681","ocarina"],"685":["n03841143","odometer"],"686":["n03843555","oil filter"],"687":["n03854065","organ"],"688":["n03857828","oscilloscope"],"689":["n03866082","overskirt"],"690":["n03868242","oxcart"],"691":["n03868863","oxygen mask"],"692":["n03871628","packet"],"693":["n03873416","paddle"],"694":["n03874293","paddlewheel"],"695":["n03874599","padlock"],"696":["n03876231","paintbrush"],"697":["n03877472","pajama"],"698":["n03877845","palace"],"699":["n03884397","panpipe"],"700":["n03887697","paper towel"],"701":["n03888257","parachute"],"702":["n03888605","parallel bars"],"703":["n03891251","park bench"],"704":["n03891332","parking meter"],"705":["n03895866","passenger car"],"706":["n03899768","patio"],
"707":["n03902125","pay-phone"],"708":["n03903868","pedestal"],"709":["n03908618","pencil box"],"710":["n03908714","pencil sharpener"],"711":["n03916031","perfume"],"712":["n03920288","Petri dish"],"713":["n03924679","photocopier"],"714":["n03929660","pick"],"715":["n03929855","pickelhaube"],"716":["n03930313","picket fence"],"717":["n03930630","pickup"],"718":["n03933933","pier"],"719":["n03935335","piggy bank"],"720":["n03937543","pill bottle"],"721":["n03938244","pillow"],"722":["n03942813","ping-pong ball"],"723":["n03944341","pinwheel"],"724":["n03947888","pirate"],"725":["n03950228","pitcher"],"726":["n03954731","plane"],"727":["n03956157","planetarium"],"728":["n03958227","plastic bag"],"729":["n03961711","plate rack"],"730":["n03967562","plow"],"731":["n03970156","plunger"],"732":["n03976467","Polaroid camera"],"733":["n03976657","pole"],"734":["n03977966","police van"],"735":["n03980874","poncho"],"736":["n03982430","pool table"],"737":["n03983396","pop bottle"],"738":["n03991062","pot"],"739":["n03992509","potter's wheel"],"740":["n03995372","power drill"],"741":["n03998194","prayer rug"],"742":["n04004767","printer"],"743":["n04005630","prison"],"744":["n04008634","projectile"],"745":["n04009552","projector"],"746":["n04019541","puck"],"747":["n04023962","punching bag"],"748":["n04026417","purse"],"749":["n04033901","quill"],"750":["n04033995","quilt"],"751":["n04037443","racer"],"752":["n04039381","racket"],"753":["n04040759","radiator"],"754":["n04041544","radio"],"755":["n04044716","radio telescope"],"756":["n04049303","rain barrel"],"757":["n04065272","recreational vehicle"],"758":["n04067472","reel"],"759":["n04069434","reflex camera"],"760":["n04070727","refrigerator"],"761":["n04074963","remote control"],"762":["n04081281","restaurant"],"763":["n04086273","revolver"],"764":["n04090263","rifle"],"765":["n04099969","rocking chair"],"766":["n04111531","rotisserie"],"767":["n04116512","rubber eraser"],"768":["n04118538","rugby ball"],"769":["n04118776","rule"],"770":["n04120489","running shoe"],"771":["n04125021","safe"],"772":["n04127249","safety pin"],"773":["n04131690","saltshaker"],"774":["n04133789","sandal"],"775":["n04136333","sarong"],"776":["n04141076","sax"],"777":["n04141327","scabbard"],"778":["n04141975","scale"],"779":["n04146614","school bus"],"780":["n04147183","schooner"],"781":["n04149813","scoreboard"],"782":["n04152593","screen"],"783":["n04153751","screw"],"784":["n04154565","screwdriver"],"785":["n04162706","seat belt"],"786":["n04179913","sewing machine"],"787":["n04192698","shield"],"788":["n04200800","shoe shop"],"789":["n04201297","shoji"],"790":["n04204238","shopping basket"],"791":["n04204347","shopping cart"],"792":["n04208210","shovel"],"793":["n04209133","shower cap"],"794":["n04209239","shower curtain"],"795":["n04228054","ski"],"796":["n04229816","ski mask"],"797":["n04235860","sleeping bag"],"798":["n04238763","slide rule"],"799":["n04239074","sliding door"],"800":["n04243546","slot"],"801":["n04251144","snorkel"],"802":["n04252077","snowmobile"],"803":["n04252225","snowplow"],"804":["n04254120","soap dispenser"],"805":["n04254680","soccer ball"],"806":["n04254777","sock"],"807":["n04258138","solar dish"],"808":["n04259630","sombrero"],"809":["n04263257","soup bowl"],"810":["n04264628","space bar"],
"811":["n04265275","space heater"],"812":["n04266014","space shuttle"],"813":["n04270147","spatula"],"814":["n04273569","speedboat"],"815":["n04275548","spider web"],"816":["n04277352","spindle"],"817":["n04285008","sports car"],"818":["n04286575","spotlight"],"819":["n04296562","stage"],"820":["n04310018","steam locomotive"],"821":["n04311004","steel arch bridge"],"822":["n04311174","steel drum"],"823":["n04317175","stethoscope"],"824":["n04325704","stole"],"825":["n04326547","stone wall"],"826":["n04328186","stopwatch"],"827":["n04330267","stove"],"828":["n04332243","strainer"],"829":["n04335435","streetcar"],"830":["n04336792","stretcher"],"831":["n04344873","studio couch"],"832":["n04346328","stupa"],"833":["n04347754","submarine"],"834":["n04350905","suit"],"835":["n04355338","sundial"],"836":["n04355933","sunglass"],"837":["n04356056","sunglasses"],"838":["n04357314","sunscreen"],"839":["n04366367","suspension bridge"],"840":["n04367480","swab"],"841":["n04370456","sweatshirt"],"842":["n04371430","swimming trunks"],"843":["n04371774","swing"],"844":["n04372370","switch"],"845":["n04376876","syringe"],"846":["n04380533","table lamp"],"847":["n04389033","tank"],"848":["n04392985","tape player"],"849":["n04398044","teapot"],"850":["n04399382","teddy"],"851":["n04404412","television"],"852":["n04409515","tennis ball"],"853":["n04417672","thatch"],"854":["n04418357","theater curtain"],"855":["n04423845","thimble"],"856":["n04428191","thresher"],"857":["n04429376","throne"],"858":["n04435653","tile roof"],"859":["n04442312","toaster"],"860":["n04443257","tobacco shop"],"861":["n04447861","toilet seat"],"862":["n04456115","torch"],"863":["n04458633","totem pole"],"864":["n04461696","tow truck"],"865":["n04462240","toyshop"],"866":["n04465501","tractor"],"867":["n04467665","trailer truck"],"868":["n04476259","tray"],"869":["n04479046","trench coat"],"870":["n04482393","tricycle"],"871":["n04483307","trimaran"],"872":["n04485082","tripod"],"873":["n04486054","triumphal arch"],"874":["n04487081","trolleybus"],"875":["n04487394","trombone"],"876":["n04493381","tub"],"877":["n04501370","turnstile"],"878":["n04505470","typewriter keyboard"],"879":["n04507155","umbrella"],"880":["n04509417","unicycle"],"881":["n04515003","upright"],"882":["n04517823","vacuum"],"883":["n04522168","vase"],"884":["n04523525","vault"],"885":["n04525038","velvet"],"886":["n04525305","vending machine"],"887":["n04532106","vestment"],"888":["n04532670","viaduct"],"889":["n04536866","violin"],"890":["n04540053","volleyball"],"891":["n04542943","waffle iron"],"892":["n04548280","wall clock"],"893":["n04548362","wallet"],"894":["n04550184","wardrobe"],"895":["n04552348","warplane"],"896":["n04553703","washbasin"],"897":["n04554684","washer"],"898":["n04557648","water bottle"],"899":["n04560804","water jug"],"900":["n04562935","water tower"],"901":["n04579145","whiskey jug"],"902":["n04579432","whistle"],"903":["n04584207","wig"],"904":["n04589890","window screen"],"905":["n04590129","window shade"],"906":["n04591157","Windsor tie"],"907":["n04591713","wine bottle"],"908":["n04592741","wing"],"909":["n04596742","wok"],"910":["n04597913","wooden spoon"],"911":["n04599235","wool"],"912":["n04604644","worm fence"],"913":["n04606251","wreck"],"914":["n04612504","yawl"],
"915":["n04613696","yurt"],"916":["n06359193","web site"],"917":["n06596364","comic book"],"918":["n06785654","crossword puzzle"],"919":["n06794110","street sign"],"920":["n06874185","traffic light"],"921":["n07248320","book jacket"],"922":["n07565083","menu"],"923":["n07579787","plate"],"924":["n07583066","guacamole"],"925":["n07584110","consomme"],"926":["n07590611","hot pot"],"927":["n07613480","trifle"],"928":["n07614500","ice cream"],"929":["n07615774","ice lolly"],"930":["n07684084","French loaf"],"931":["n07693725","bagel"],"932":["n07695742","pretzel"],"933":["n07697313","cheeseburger"],"934":["n07697537","hotdog"],"935":["n07711569","mashed potato"],"936":["n07714571","head cabbage"],"937":["n07714990","broccoli"],"938":["n07715103","cauliflower"],"939":["n07716358","zucchini"],"940":["n07716906","spaghetti squash"],"941":["n07717410","acorn squash"],"942":["n07717556","butternut squash"],"943":["n07718472","cucumber"],"944":["n07718747","artichoke"],"945":["n07720875","bell pepper"],"946":["n07730033","cardoon"],"947":["n07734744","mushroom"],"948":["n07742313","Granny Smith"],"949":["n07745940","strawberry"],"950":["n07747607","orange"],"951":["n07749582","lemon"],"952":["n07753113","fig"],"953":["n07753275","pineapple"],"954":["n07753592","banana"],"955":["n07754684","jackfruit"],"956":["n07760859","custard apple"],"957":["n07768694","pomegranate"],"958":["n07802026","hay"],"959":["n07831146","carbonara"],"960":["n07836838","chocolate sauce"],"961":["n07860988","dough"],"962":["n07871810","meat loaf"],"963":["n07873807","pizza"],"964":["n07875152","potpie"],"965":["n07880968","burrito"],"966":["n07892512","red wine"],"967":["n07920052","espresso"],"968":["n07930864","cup"],"969":["n07932039","eggnog"],"970":["n09193705","alp"],"971":["n09229709","bubble"],"972":["n09246464","cliff"],"973":["n09256479","coral reef"],"974":["n09288635","geyser"],"975":["n09332890","lakeside"],"976":["n09399592","promontory"],"977":["n09421951","sandbar"],"978":["n09428293","seashore"],"979":["n09468604","valley"],"980":["n09472597","volcano"],"981":["n09835506","ballplayer"],"982":["n10148035","groom"],"983":["n10565667","scuba diver"],"984":["n11879895","rapeseed"],"985":["n11939491","daisy"],"986":["n12057211","yellow lady's slipper"],"987":["n12144580","corn"],"988":["n12267677","acorn"],"989":["n12620546","hip"],"990":["n12768682","buckeye"],"991":["n12985857","coral fungus"],"992":["n12998815","agaric"],"993":["n13037406","gyromitra"],"994":["n13040303","stinkhorn"],"995":["n13044778","earthstar"],"996":["n13052670","hen-of-the-woods"],"997":["n13054560","bolete"],"998":["n13133613","ear"],"999":["n15075141","toilet tissue"]}
Replenish
The above content is fine, but a little complicated. I recently read some information and would like to add some additional information.
Model download
Regarding model download, for the models provided by Google, you can download them locally and then load them. But if it is a Google model, it is usually built into TensorFlow. You can use it directly, and the corresponding file will be automatically downloaded when using it. This piece was not found in the official documentation when I first learned it, and it was a bit complicated. The above code can be optimized into the following content
Optimized
# 导入tensorflow 和科学计算库
import tensorflow as tf
# 字体属性
from matplotlib.font_manager import FontProperties
# matplotlib是用于绘制图表和可视化数据的库
import matplotlib.pylab as plt
# 加载内置的模型
model = tf.keras.applications.mobilenet_v2.MobileNetV2()
# 加载并处理图片
# 1、加载本地图片，并将其处理为224*224
image = tf.keras.preprocessing.image.load_img('./pics/dog.jpg',target_size=(224,224))
# 2、转成数组
image = tf.keras.preprocessing.image.img_to_array(image)
print("图片形状：",image.shape)
# 3、扩展维度
image = image.reshape(1,image.shape[0],image.shape[1],image.shape[2])
# 或者 image = np.expand_dims(image, axis=0)
# 4、处理输入
image = tf.keras.applications.mobilenet_v2.preprocess_input(image)
# 5、预测
predictions = model.predict(image)
print("预测结果：",predictions)
# 6、解码，获取可能性最大的
label = tf.keras.applications.mobilenet_v2.decode_predictions(predictions,top=1)
print("解码结果：",label)
# 7、可视化显示
font = FontProperties()
font.set_family('Microsoft YaHei')
plt.figure() # 创建图像窗口
plt.xticks([])
plt.yticks([])
plt.grid(False) # 取消网格线
plt.imshow(image[0]) # 显示图片
plt.xlabel(label[0],fontproperties=font)
plt.show() # 显示图形窗口
You can see from the above code that it is more convenient to directly use the official built-in model. In addition, model-related files will be automatically downloaded and saved in the .keras/models/ folder in the user's home directory. If the folder does not exist, it will be created automatically. Model files will be downloaded into this directory and stored in .h5files.
Fine-tune the model
The MobileNetV2 documentation says that you can fine-tune the model according to your own needs, so how should you do it?
Official tutorial:Using TensorFlow Hub for transfer learning
This tutorial also happens to use the MobileNetV2 model. We can refer to the official tutorial to fine-tune the model.
use
The tutorial introduces how to use MobileNetV2, here we can learn some other knowledge.
Predicting web images
import PIL.Image as Image
# 1、mobilenet需要的图片尺寸是 224 * 224
#image = tf.keras.preprocessing.image.load_img('pics/dog.png',target_size=(224,224))
image = tf.keras.utils.get_file('bird.jpg','https://scpic.chinaz.net/files/default/imgs/2023-08-29/7dc085b6d3291303.jpg')
image = Image.open(image).resize((224,224))
Please note here that after you change the network image address, you must modify the corresponding name, otherwise the last image will be loaded.
Get tags
I used to worry about inconsistent data sets, here the official provides us with a way
labels_path = tf.keras.utils.get_file('ImageNetLabels.txt','https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt')
imagenet_labels = np.array(open(labels_path).read().splitlines())
This tag is plain text and requires scientific Internet access to access it.
transfer learning
How to do it? Official document introduction
1. We need a data set, and the document has been prepared for us (first use the official data set, and then study how to make your own data set)
2. From < /span> 3. Retrain the top (last) layer to identify classes in a custom dataset Download a pre-trained model, which we already haveTensorFlow Hub
data, flower data
import pathlib
# 下载并解压到当前目录
data_file = tf.keras.utils.get_file(
'flower_photos.tgz',
'https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz',
cache_dir='.',
extract=True)
data_root = pathlib.Path(data_file).with_suffix('')
The above is the official document method. Here we download it directly through the browser and extract it to the current folder (requires scientific Internet access)
The file is still relatively large
After looking at the pictures, they are all ordinary pictures. This will make it easier for us to organize the training set data later. Of course, it is best not to make the pictures too big.
Load data and divide the data set
# 32张图片为一个批次，尺寸设置为224*224
batch_size = 32
img_height = 224
img_width = 224
# 加载图像数据集，并将其分割为训练集和验证集，验证集比例为20%
train_ds = tf.keras.utils.image_dataset_from_directory(
'flower_photos', # 目录
validation_split=0.2, # 验证集占20%
subset="training", # 将数据集划分为训练集
seed= 123, # 随机种子，用于数据集随机划分
image_size= (img_width,img_height) , # 调整图像大小
batch_size= batch_size # 每个批次中包含的图像数量
)
# 验证集
val_ds = tf.keras.utils.image_dataset_from_directory(
'flower_photos', # 目录
validation_split=0.2, # 验证集占20%
subset="validation", # 将数据集划分为验证集
seed= 123, # 随机种子，用于数据集随机划分
image_size= (img_width,img_height) , # 调整图像大小
batch_size= batch_size # 每个批次中包含的图像数量
)
After execution, you can see that the console outputs the following information. 3670 pictures were found, and these pictures were classified into 5 categories, of which 2936 pictures were used for training. It is classified into 5 categories here, just for there are 5 different pictures in the directory
Get flower species
# 花卉种类
class_names = np.array(train_ds.class_names)
print("花卉种类：",class_names)
Correctly corresponds to the names of those 5 folders
Normalize the image data of the training set and verification set
Normalized The purpose has been explained in another article and will not be discussed here
# 归一化
normalization_layer = tf.keras.layers.Rescaling(1./255) # 创建了一个Rescaling层，将像素值缩放到0到1之间 。 1./255是 1/255保留小数，差点没看懂
train_ds = train_ds.map(lambda x,y:(normalization_layer(x),y))
val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y))
Caching and prefetching data sets
# 使用缓冲预取，避免产生I/O阻塞
AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
# 验证数据是否成功加载和处理
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break
The output result is consistent with the official document, you can continue
(32, 224, 224, 3) means that there are 32 images in each batch, the size is224*224, and the number of channels of the image is 3 (RGB channel)
Run the classifier on a batch of images to make predictions
# 对一批图片运行分类器，进行预测
result_batch = model.predict(train_ds)
# 加载映射文件，这里我将其下载到了本地
# 文件下载地址：https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt
imagenet_labels = np.array(open('labels.txt').read().splitlines())
# 在给定的张量中找到沿指定轴的最大值的索引
predict_class_names = imagenet_labels[tf.math.argmax(result_batch, axis=-1)]
print("预测类别：",predict_class_names)
# 绘制出预测与图片
plt.figure(figsize=(10,9))
plt.subplots_adjust(hspace=0.5)
for n in range(30):
plt.subplot(6,5,n+1)
plt.imshow(image_batch[n])
plt.title(predict_class_names[n])
plt.axis('off')
_ = plt.suptitle("ImageNet predictions")
plt.show()
It is consistent with the official operation effect and you can continue to proceed.
Download the headless mode
The model used before was used for classification. Now you need to download the feature extraction model according to the official tutorial. Classification modelimagenet_mobilenet_v2_140_224_classification_5 The corresponding feature extraction model download address is:https://tfhub.dev/google/imagenet /mobilenet_v2_140_224/feature_vector/5
Download and extract it into the project and use it
# 加载特征提取器
feature_extractor_layer = hub.KerasLayer(
'imagenet_mobilenet_v2_140_224_feature_vector_5', # 预训练模型
input_shape=(224,224,3), # 指定图像输入的高度、宽度和通道数
trainable=False #训练过程中不更新特征提取器的权重
)
# 特征提取器为每个图像返回一个 1280 长的向量（在此示例中，图像批大小仍为 32）
feature_batch = feature_extractor_layer(image_batch)
print("特征批次形状：",feature_batch.shape)
The problem is not big, it is caused by the different models used.
Additional classification header
# 附加分类头
new_model = tf.keras.Sequential([
feature_extractor_layer,
tf.keras.layers.Dense(len(class_names),activation='softmax') # 指定输出分类，这里的花是5类
])
Can be executed normally, continue
Training model
Now that the features have been extracted, the next step is to train the model.
# 开始训练，暂时只训练10轮。history记录了训练过程中的各项指标，便于后续分析和可视化
history = new_model.fit(
train_ds, # 训练数据集
validation_data=val_ds, # 验证数据集，用于在训练过程中监控模型的性能
epochs=10, # 训练的总轮次
callbacks=tensorboard_callback # 回调函数，用于在训练过程中执行特定操作，比如记录日志
)
As you can see from the picture below, after 10 rounds of training, the accuracy is getting higher and higher
The training log can also be seen in the folder:
Also run in the console: tensorboard --logdir logs/fit Start TensorBoard
The console will return a URL where you can see how the metric changes with each epoch, and track other scalars value
Check forecast
We’re basically done here, now let’s make a simple prediction
predicted_batch = new_model.predict(image_batch)
predicted_id = tf.math.argmax(predicted_batch, axis=-1)
predicted_label_batch = class_names[predicted_id]
print("花卉种类：",predicted_label_batch)
plt.figure(figsize=(10,9))
plt.subplots_adjust(hspace=0.5)
for n in range(30):
plt.subplot(6,5,n+1)
plt.imshow(image_batch[n])
plt.title(predicted_label_batch[n].title())
plt.axis('off')
_ = plt.suptitle("Model predictions")
plt.show()
The running results are consistent with the official documents
Export and reload the model
After the model is trained, it can be exported asSavedModel for future reuse.
SavedModel is a model saving format in TensorFlow. It is a standardized format for saving trained models to disk and making them easy to load and use. SavedModel can save the model's architecture, weight parameters, optimizer status, and other model-related settings.
SavedModel provides a cross-platform and cross-language model saving and loading method that can be used in different TensorFlow versions and different programming languages. This allows us to move models from one environment to another, or share models with other TensorFlow applications.
When exporting a model, you can use thetf.saved_model.save() function to save the model in SavedModel format. This function accepts the model object and the saving path as parameters, and saves the model-related information to the folder under the specified path. For example:
tf.saved_model.save(model, export_path)
Among them, model is the model object you want to save, and export_path is the saving path.
After exporting the model, you can use thetf.saved_model.load() function elsewhere to load the SavedModel and use the loaded model for prediction or other operations.
Export
# 导出训练好的模型
export_path = '/tmp/saved_models/flower_model'
new_model.save(export_path)
Complete code
# 导入tensorflow 和科学计算库
import tensorflow as tf
import numpy as np
# tensorflow-hub是一个TensorFlow库的扩展，它提供了一个简单的接口，用于重用已经训练好的机器学习模型的部分
import tensorflow_hub as hub
# 字体属性
from matplotlib.font_manager import FontProperties
# matplotlib是用于绘制图表和可视化数据的库
import matplotlib.pylab as plt
import datetime
# 导入模型
# 不能直接加载模型文件，需要加载器目录
model = tf.keras.Sequential([
hub.KerasLayer('imagenet_mobilenet_v2_140_224_classification_5')
])
# 32张图片为一个批次，尺寸设置为224*224
batch_size = 32
img_height = 224
img_width = 224
# 加载图像数据集，并将其分割为训练集和验证集，验证集比例为20%
train_ds = tf.keras.utils.image_dataset_from_directory(
'flower_photos', # 目录
validation_split=0.2, # 验证集占20%
subset="training", # 将数据集划分为训练集
seed= 123, # 随机种子，用于数据集随机划分
image_size= (img_width,img_height) , # 调整图像大小
batch_size= batch_size # 每个批次中包含的图像数量
)
# 验证集
val_ds = tf.keras.utils.image_dataset_from_directory(
'flower_photos', # 目录
validation_split=0.2, # 验证集占20%
subset="validation", # 将数据集划分为验证集
seed= 123, # 随机种子，用于数据集随机划分
image_size= (img_width,img_height) , # 调整图像大小
batch_size= batch_size # 每个批次中包含的图像数量
)
# 花卉种类
class_names = np.array(train_ds.class_names)
print("花卉种类：",class_names)
# 归一化
normalization_layer = tf.keras.layers.Rescaling(1./255) # 创建了一个Rescaling层，将像素值缩放到0到1之间 。 1./255是 1/255保留小数，差点没看懂
train_ds = train_ds.map(lambda x,y:(normalization_layer(x),y))
val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y))
# 使用缓冲预取，避免产生I/O阻塞
AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
# 验证数据是否成功加载和处理
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break
# 对一批图片运行分类器，进行预测
result_batch = model.predict(train_ds)
# 加载映射文件，这里我将其下载到了本地
imagenet_labels = np.array(open('labels.txt').read().splitlines())
# 在给定的张量中找到沿指定轴的最大值的索引
predict_class_names = imagenet_labels[tf.math.argmax(result_batch, axis=-1)]
print("预测类别：",predict_class_names)
# 绘制出预测与图片
# plt.figure(figsize=(10,9))
# plt.subplots_adjust(hspace=0.5)
# for n in range(30):
# plt.subplot(6,5,n+1)
# plt.imshow(image_batch[n])
# plt.title(predict_class_names[n])
# plt.axis('off')
# _ = plt.suptitle("ImageNet predictions")
# plt.show()
# 加载特征提取器
feature_extractor_layer = hub.KerasLayer(
'imagenet_mobilenet_v2_140_224_feature_vector_5', # 预训练模型
input_shape=(224,224,3), # 指定图像输入的高度、宽度和通道数
trainable=False #训练过程中不更新特征提取器的权重
)
# 特征提取器为每个图像返回一个 1280 长的向量（在此示例中，图像批大小仍为 32）
feature_batch = feature_extractor_layer(image_batch)
print("特征批次形状：",feature_batch.shape)
# 附加分类头
new_model = tf.keras.Sequential([
feature_extractor_layer,
tf.keras.layers.Dense(len(class_names),activation='softmax') # 指定输出分类，这里的花是5类
])
# 训练模型
new_model.compile(
optimizer=tf.keras.optimizers.Adam(), # 使用Adam优化器作为优化算法
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), # 使用SparseCategoricalCrossentropy作为损失函数
metrics=['acc'] # 使用准确率作为评估指标
)
# 训练日志
log_dir = "logs/fit/" + datetime.datetime.now().strftime('%Y%m%d-%H%M%S')
# 用于在训练过程中收集模型指标和摘要数据，并将其写入TensorBoard日志文件中
tensorboard_callback = tf.keras.callbacks.TensorBoard(
log_dir= log_dir,
histogram_freq=1
)
# 开始训练，暂时只训练10轮。history记录了训练过程中的各项指标，便于后续分析和可视化
history = new_model.fit(
train_ds, # 训练数据集
validation_data=val_ds, # 验证数据集，用于在训练过程中监控模型的性能
epochs=10, # 训练的总轮次
callbacks=tensorboard_callback # 回调函数，用于在训练过程中执行特定操作，比如记录日志
)
# 简单预测
# predicted_batch = new_model.predict(image_batch)
# predicted_id = tf.math.argmax(predicted_batch, axis=-1)
# predicted_label_batch = class_names[predicted_id]
# print("花卉种类：",predicted_label_batch)
# plt.figure(figsize=(10,9))
# plt.subplots_adjust(hspace=0.5)
# for n in range(30):
# plt.subplot(6,5,n+1)
# plt.imshow(image_batch[n])
# plt.title(predicted_label_batch[n].title())
# plt.axis('off')
# _ = plt.suptitle("Model predictions")
# plt.show()
# 导出训练好的模型
export_path = 'tmp/saved_models/flower_model'
new_model.save(export_path)
Use the fine-tuned model
The model has been saved locally, now let’s use it
# 导入tensorflow 和科学计算库
import tensorflow as tf
import numpy as np
# matplotlib是用于绘制图表和可视化数据的库
import matplotlib.pylab as plt
import PIL.Image as Image
from matplotlib.font_manager import FontProperties
# 导入模型
model = tf.keras.models.load_model('tmp/saved_models/flower_model')
# 预处理输入数据
image = tf.keras.utils.get_file('sunflower.jpg','https://scpic.chinaz.net/files/pic/pic9/202006/bpic20492.jpg')
image = Image.open(image).resize((224,224))
image = tf.keras.preprocessing.image.img_to_array(image)
image = np.expand_dims(image, axis=0)
image = tf.keras.applications.mobilenet_v2.preprocess_input(image)
# 预测
predictions = model.predict(image)
predicted_index = np.argmax(predictions)
print("索引是：",predicted_index)
class_names= ['雏菊','蒲公英' ,'玫瑰' ,'向日葵' ,'郁金香']
# 可视化显示
font = FontProperties()
font.set_family('Microsoft YaHei')
plt.figure() # 创建图像窗口
plt.xticks([])
plt.yticks([])
plt.grid(False) # 取消网格线
plt.imshow(image[0]) # 显示图片
plt.xlabel(class_names[predicted_index] ,fontproperties=font)
plt.show() # 显示图形窗口
very successful
Use your own data to train the model
The above code is very successful, but the data used is officially provided. Here we replace the data set with our own based on the above tutorial.
I simply prepared 45 cat pictures here, 15 white cats, 15 black cats, and 15 black and white cats. The amount of data is not enough. If it is not enough, you can simply try the effect
Data
Training and exporting model
The accuracy is a bit high and I feel a bit panicked
use
Can you still accept it?
