{"id":102,"date":"2018-01-14T20:00:12","date_gmt":"2018-01-14T11:00:12","guid":{"rendered":"http:\/\/staka.jp\/wordpress\/?p=102"},"modified":"2018-01-14T20:00:12","modified_gmt":"2018-01-14T11:00:12","slug":"%e6%a4%8d%e7%89%a9%e3%81%ae%e7%97%85%e6%b0%97%e3%82%92deep-learning%e3%81%a7%e5%88%a4%e5%88%a5-%e3%80%80%e3%83%a2%e3%83%87%e3%83%ab%e6%a7%8b%e7%af%89%e7%b7%a8%ef%bc%882-3%ef%bc%89","status":"publish","type":"post","link":"https:\/\/staka.jp\/wordpress\/?p=102","title":{"rendered":"\u30e2\u30c7\u30eb\u69cb\u7bc9\uff08keras\/tensorflow+InceptionV3+Data augmentation\uff09\u7de8\uff082\/3\uff09"},"content":{"rendered":"<p>\u524d\u56de\u304b\u3089\u5f15\u304d\u7d9a\u304d\u3001<a href=\"http:\/\/www.plant-check.jp\/\">\u690d\u7269\u306e\u5199\u771f\u304b\u3089\u75c5\u6c17\u3092\u5224\u5225\u3059\u308b\u30b5\u30a4\u30c8(http:\/\/www.plant-check.jp\/)<\/a>\u3092\u4f5c\u3063\u305f\u3068\u304d\u306e\u307e\u3068\u3081\u3002<\/p>\n<h2>\u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u6d41\u308c<\/h2>\n<p>\u524d\u56de\u66f8\u3044\u305f\u3068\u304a\u308a\u3001\u4eca\u56de\u3001(1)\u690d\u7269\u306e\u8449\u304b\u305d\u308c\u4ee5\u5916\u304b\u3092\u5224\u5225\u3059\u308b2\u5024\u5206\u985e\u30e2\u30c7\u30eb\uff0b(2)\u690d\u7269\u306e\u8449\u304c\u75c5\u6c17\u304b\u5426\u304b\u3092\u5224\u5225\u3059\u308b\u591a\u5024\u5206\u985e\u30e2\u30c7\u30eb\u3092\u4f5c\u6210\u3057\u305f\u3002\u69cb\u7bc9\u65b9\u6cd5\u306f(1)\u3001(2)\u3068\u3082\u306b\u540c\u69d8\u3067keras, tensorflow\u3092\u7528\u3044\u3066\u8ee2\u79fb\u5b66\u7fd2\u3092\u884c\u3063\u305f\u3002\u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u6d41\u308c\u306f\u4e0b\u8a18\u306e\u901a\u308a\u3002\u30d0\u30ea\u30d0\u30ea\u306eDeep Learning\u306a\u306e\u3067\u3001\u4eba\u5de5\u77e5\u80fd\uff08AI\uff09\u3092\u5229\u7528\u3057\u3001\u690d\u7269\u306e\u75c5\u6c17\u5224\u5b9a\u3092\u884c\u3063\u305f\uff01\u3068\u8a00\u3063\u3066\u3044\u3044\u306f\u305a\uff08\u5927\u4e8b\u306a\u3053\u3068\u306a\u306e\u30672\u56de\u76ee\uff09\u3002<\/p>\n<ol>\n<li>\u5199\u771f\u3092\u30e9\u30d9\u30eb\u3054\u3068\uff08(1)\u306f\u300c\u690d\u7269\u306e\u8449\/\u305d\u308c\u4ee5\u5916\u300d\u3001(2)\u306f\u300c\u5065\u5eb7\u306a\u8449\/\u9ed2\u661f\u75c5\/\u3046\u3069\u3093\u7c89\u75c5\/\u305d\u306e\u4ed6\u30ab\u30d3\u7cfb\u306e\u75c5\u6c17\u300d\uff09\u3054\u3068\u306b\u5225\u306e\u30c7\u30a3\u30ec\u30af\u30c8\u30ea\u306b\u683c\u7d0d\u3059\u308b\u3002<\/li>\n<li>keras\u3092\u4f7f\u3063\u3066ImageNet\u3092\u5b66\u7fd2\u3057\u305fInception V3\u3092\u30ed\u30fc\u30c9\u3059\u308b\u3002<\/li>\n<li>\u30ed\u30fc\u30c9\u3057\u305f\u30e2\u30c7\u30eb\u306e\u4e00\u90e8\uff08\u4eca\u56de\u306f250\u5c64\u4ee5\u964d\uff09\u3092\u5b66\u7fd2\u53ef\u80fd\u306b\u8a2d\u5b9a\u3059\u308b\u3002<\/li>\n<li>\u554f\u984c\uff082\u5024 or \u591a\u5024\uff09\u306b\u5fdc\u3058\u3066\u3001Inecption V3\u306e\u5f8c\u6bb5\u306e\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u8a2d\u5b9a\u3059\u308b\u3002<\/li>\n<li>\u5199\u771f\u30c7\u30fc\u30bf\u3092\u5b66\u7fd2\u7528\u3068\u691c\u8a3c\u306b\u5206\u3051\u308b\u3002<\/li>\n<li>\u5199\u771f\u30c7\u30fc\u30bf\u3092data augmentation\u3059\u308b\u3088\u3046\u8a2d\u5b9a\u3059\u308b\u3002<\/li>\n<li>\u5b66\u7fd2\u3001\u691c\u8a3c\u3059\u308b\u3002<\/li>\n<\/ol>\n<h2>\u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u30b3\u30fc\u30c9<\/h2>\n<p>\u4e0a\u8a181.\uff5e7.\u306e\u30d5\u30ed\u30fc\u306fkeras (+ tensorflow\uff09\u3092\u7528\u3044\u308b\u3068\u7c21\u5358\u306b\u5b9f\u88c5\u3067\u304d\u308b\u30022\u5024\u5206\u985e\u30e2\u30c7\u30eb\u306b\u304a\u3051\u308b\u30012.\uff5e4.\u306f\u4e0b\u8a18\u306e\u3088\u3046\u306b\u5b9f\u88c5\u53ef\u80fd\u3002\u591a\u5024\u306e\u5834\u5408\u306f\u6700\u7d42\u5c64\u3092\u300cpredictions = Dense(\u30af\u30e9\u30b9\u6570, activation=&#8221;softmax&#8221;)(x)\u300d\u3066\u306a\u611f\u3058\u306b\u5909\u3048\u3066\u300closs = &#8220;categorical_crossentropy&#8221;\u300d\u3068\u3059\u308c\u3070\u3088\u3044\u3002\u3056\u3063\u304f\u308a\u3044\u3046\u3068\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u524d\u6bb5\u3068\u3057\u3066\u5b66\u7fd2\u6e08\u307f\u306e\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u7528\u3044\u7279\u5fb4\u91cf\u62bd\u51fa\u7b49\u3092\u518d\u5229\u7528\uff08\uff1d\u8ee2\u79fb\u5b66\u7fd2\uff09\u3057\u3001\u672c\u4ef6\u3067\u5fc5\u8981\u306a\u5206\u985e\u3092\u884c\u3046\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u8ffd\u52a0\u3057\u3066\u3044\u308b\u3002<\/p>\n<pre>base_model = InceptionV3(weights='imagenet', include_top=False)\nx = base_model.output\nx = GlobalAveragePooling2D()(x)\nx = Dense(1024, activation=\"relu\")(x)\nx = Dropout(0.5)(x)\nx = Dense(256, activation=\"relu\")(x)\npredictions = Dense(1, activation=\"sigmoid\")(x)\nmodel = Model(inputs=base_model.input, outputs=predictions)\nfor layer in model.layers[:249]:\n   layer.trainable = False\nfor layer in model.layers[249:]:\n   layer.trainable = True\nmodel.compile(loss = \"binary_crossentropy\", optimizer = optimizers.SGD(lr=0.0001, momentum=0.9), metrics=[\"accuracy\"])\n<\/pre>\n<p>Data augmentation\u306f\u753b\u50cf\u3092\u5909\u5f62\uff08\u56de\u8ee2\u3001\u79fb\u52d5\u3001\u7e2e\u5c0f\u3001\u62e1\u5927\u306a\u3069\u306a\u3069\uff09\u3055\u305b\u306a\u304c\u3089\u30c7\u30fc\u30bf\u3092\u5897\u3084\u3057\u3066\u5b66\u7fd2\u3059\u308b\u65b9\u6cd5\u3067\u3042\u308b\u3002\u753b\u50cf\u3092\u5909\u5f62\u3055\u305b\u308b\u3053\u3068\u3067\u30011\u3064\u306e\u753b\u50cf\u304b\u3089\u8907\u6570\u306e\u30d1\u30bf\u30fc\u30f3\u3092\u751f\u307f\u51fa\u3059\u3002\u65e5\u672c\u8a9e\u3060\u3068\u30c7\u30fc\u30bf\u62e1\u5f35\u3068\u304b\u547c\u3070\u308c\u3066\u3044\u308b\uff08\u306f\u305a\uff09\u3002\u5199\u771f\u306e\u64ae\u3089\u308c\u65b9\u306b\u4f9d\u5b58\u3059\u308b\u5dee\u7570\u7b49\u304c\u5438\u53ce\u3067\u304d\u308b\u306e\u3067\u3001\u672a\u77e5\u30c7\u30fc\u30bf\u306b\u5bfe\u3059\u308b\u6027\u80fd\u306e\u5411\u4e0a\u306b\u52b9\u679c\u304c\u3042\u308b\uff08\u3068\u79c1\u306f\u601d\u3063\u3066\u3044\u308b\uff09\u3002\u30e2\u30c7\u30eb\u9069\u7528\u6642\u306b\u3082\u5909\u5f62\u3055\u305b\u306a\u304c\u3089\u4f55\u30d1\u30bf\u30fc\u30f3\u304b\u9069\u7528\u3057\u3001\u305d\u306e\u5e73\u5747\u3092\u53d6\u308b\u3068\u6027\u80fd\u5411\u4e0a\u52b9\u679c\u304c\u3042\u308b\uff08\u3053\u3068\u3082\u3042\u308b\uff09\u304c\u3001\u3053\u3063\u3061\u3092\u30c7\u30fc\u30bf\u62e1\u5f35\u3068\u8a00\u3046\u304b\u306f\u3088\u304f\u308f\u304b\u3089\u306a\u3044\u3002<br \/>\nkeras\u3060\u3068\u300cImageDataGenerator\u300d\u3092\u4f7f\u3063\u3066\u7c21\u5358\u306b\u66f8\u3051\u308b\u3002\u591a\u5024\u306e\u5834\u5408\u306f\u300cclass_mode = &#8220;categorical&#8221;\u300d\u306b\u5909\u66f4\u3059\u308c\u3070\u3088\u3044\u3002\u5b66\u7fd2\u30c7\u30fc\u30bf\u3068\u691c\u8a3c\u30c7\u30fc\u30bf\u306esplit\u3092\u4e8b\u524d\u306b\u884c\u3063\u3066\u3044\u308c\u3070\u3001\u540c\u3058\u3088\u3046\u306btest_generator\u3092\u304b\u3051\u308b\u3002<\/p>\n<pre>train_datagen = ImageDataGenerator(\n    rescale = 1.\/255,\n    horizontal_flip = True,\n    fill_mode = \"nearest\",\n    zoom_range = 0.3,\n    width_shift_range = 0.3,\n    height_shift_range=0.3,\n    rotation_range=90)\ntrain_generator = train_datagen.flow_from_directory(\n    train_data_dir,\n    target_size = (img_height, img_width),\n    batch_size = batch_size,\n    class_mode = \"binary\")\n<\/pre>\n<p>\u5b66\u7fd2\u306f\u4e0b\u8a18\u306e\u3088\u3046\u306b\u884c\u3048\u3070\u3088\u304f\u3001\u30c1\u30a7\u30c3\u30af\u30dd\u30a4\u30f3\u30c8\u3054\u3068\u306b\u30e2\u30c7\u30eb\u304c\u4fdd\u5b58\u3055\u308c\u3001\u7cbe\u5ea6\u306b\u5fdc\u3058\u3066\u81ea\u52d5\u3067\u30b9\u30c8\u30c3\u30d7\u3055\u308c\u308b\u3002nvidia-docker\u306e\u30b3\u30f3\u30c6\u30ca\u3067GPU\u7248\u306etensorflow\u3092\u5165\u308c\u3066\u3044\u308c\u3070\u3001GPU\u3092\u7528\u3044\u305f\u5b66\u7fd2\u304c\u884c\u308f\u308c\u308b\u3002\u3068\u3066\u3082\u4fbf\u5229\u3002<\/p>\n<pre>checkpoint = ModelCheckpoint(\"\/\u4fdd\u5b58\u7528\u30c7\u30a3\u30ec\u30af\u30c8\u30ea\/\u30e2\u30c7\u30eb\u540d_{epoch:02d}.h5\", monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=False, mode='auto', period=1)\nearly = EarlyStopping(monitor='val_acc', min_delta=0, patience=10, verbose=1, mode='auto')\nmodel.fit_generator(\n    train_generator,\n    samples_per_epoch = nb_train_samples,\n    epochs = epochs,\n    validation_data = validation_generator,\n    nb_val_samples = nb_validation_samples,\n    callbacks = [checkpoint, early])\n<\/pre>\n<p>\u4eca\u307e\u3067\u7d39\u4ecb\u3057\u305f\u30b3\u30fc\u30c9\u306f\u3001\u3060\u3044\u305f\u3044keras\u306esample\u30b3\u30fc\u30c9\u3092\u30d9\u30fc\u30b9\u306b\u3057\u3066\u3044\u308b\u3002\u81ea\u5206\u3067\u3084\u3063\u3066\u307f\u305f\u3044\u65b9\u306f<a href=\"https:\/\/keras.io\/ja\/\">keras\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\uff08https:\/\/keras.io\/ja\/\u5185\u306e\u30ea\u30f3\u30af\u304b\u3089\u98db\u3079\u308b\uff09<\/a>\u3092\u8aad\u3080\u306e\u304c\u304a\u52e7\u3081\u3067\u3042\u308b\u3002<\/p>\n<h2>\u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u30dd\u30a4\u30f3\u30c8<\/h2>\n<p>\u30b3\u30fc\u30c9\u81ea\u4f53\u306f\u7c21\u5358\u306b\u66f8\u3051\u3001\u5b9f\u884c\u3057\u3066\u307f\u308b\u3068\u691c\u8a3c\u30c7\u30fc\u30bf\u3067\u306e\u7cbe\u5ea6\u3082\u826f\u3044\u3068\u898b\u3048\u308b\u7d50\u679c\u304c\u5f97\u3089\u308c\u308b\u3002\u304c\u3001\u5b9f\u969b\u306b\u91cd\u8981\u306a\u306e\u306f\u300c\u672a\u77e5\u30c7\u30fc\u30bf\u306b\u5bfe\u3059\u308b\u4e88\u6e2c\u6027\u80fd\u300d\u3067\u3001\u601d\u3063\u305f\u3068\u304a\u308a\u306e\u7d50\u679c\u306b\u306a\u3089\u306a\u3044\u3053\u3068\u304c\u3042\u308b\u3002\u3053\u308c\u306f\u5b66\u7fd2\u30fb\u691c\u8a3c\u30c7\u30fc\u30bf\u306e\u95a2\u9023\u304c\u5f37\u3059\u304e\u308b\u304b\u3089\uff08=\u72ec\u7acb\u3058\u3083\u306a\u3044\u304b\u3089\uff09\u3067\u3001\u5b66\u7fd2\u7d50\u679c\u3092\u8a55\u4fa1\u3059\u308b\u4e0a\u3067\u91cd\u8981\u306a\u30dd\u30a4\u30f3\u30c8\u3068\u306a\u308b\u3002\u3053\u306e\u624b\u306e\u554f\u984c\u3078\u306e\u5bfe\u5fdc\u306f\u3068\u3063\u3066\u3082\u96e3\u3057\u3044\u3002\u79c1\u306f\u4eba\u5de5\u77e5\u80fd\u3084\u3089AI\u3084\u3089\u3068\u547c\u3070\u308c\u308b\u30e2\u30ce\u306e\u4e2d\u306b\u306f\u3001\u6b63\u3057\u304f\u8a55\u4fa1\u3055\u308c\u3066\u3044\u306a\u3044\u3001\u904e\u5927\u8a55\u4fa1\u306a\u30e4\u30c4\u3082\u591a\u3044\u3068\u601d\u3063\u3066\u3044\u308b\u3002<br \/>\nDeep Learning\u3092\u4f7f\u3063\u305f\u5834\u5408\uff08\u7279\u306b\u672c\u4ef6\u306e\u3088\u3046\u306a\u975e\u5e38\u306b\u8907\u96d1\u306a\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u7d44\u3093\u3067\u3044\u308b\u5834\u5408\uff09\u3001\u300c\u9ed2\u661f\u75c5\u306e\u7279\u5fb4\u3067\u3042\u308b\u9ed2\u306e\u6591\u70b9\u3092\u898b\u3066\u3001\u9ed2\u661f\u75c5\u3068\u5224\u65ad\u3057\u3066\u3044\u308b\u300d\u306a\u3069\u300cAI\u306e\u5224\u5b9a\u306b\u7d0d\u5f97\u611f\u304c\u3042\u308b\u304b\u300d\u306f\u308f\u304b\u308a\u306b\u304f\u3044\u3002\u691c\u8a3c\u3059\u308b\u3068\u3001\u300cAI\u306f\u30d0\u30e9\u304c\u690d\u308f\u3063\u3066\u3044\u308b\u9262\u306e\u8272\u306b\u6ce8\u76ee\u3057\u3066\u75c5\u6c17\u3060\u3068\u5224\u5225\u3059\u308b\u300d\u5834\u5408\u3082\u3042\u308b\uff08\u8a73\u3057\u304f\u306f\u5f8c\u8ff0*1\uff09\u3002<br \/>\n\u3053\u306e\u3088\u3046\u306a\u52d5\u304d\u306f\u30e2\u30c7\u30eb\u306e\u8aac\u660e\u53ef\u80fd\u6027\u306e\u6587\u8108\u3067\u3088\u304f\u8a71\u984c\u306b\u306a\u3063\u3066\u3044\u3066\u3001\u305f\u3068\u3048\u3070KDD 2016\u306e<a href=\"http:\/\/www.kdd.org\/kdd2016\/papers\/files\/rfp0573-ribeiroA.pdf\">\u300c\u201cWhy Should I Trust You?\u201dExplaining the Predictions of Any Classifier\u300d<\/a>\u306b\u8a73\u3057\u3044\u3002\u8ad6\u6587\u4e2d\u306e\u5bfe\u5fdc\u6848\u306f\u3001\u307e\u305a\u307e\u305a\u826f\u304f\u52d5\u304f\u304c\u3001\u901f\u5ea6\u9762\u306a\u3069\u4f7f\u3044\u52dd\u624b\u306f\u30a4\u30de\u30a4\u30c1\u3068\u3044\u3046\u5370\u8c61\u3092\u53d7\u3051\u305f\u3002\u8aac\u660e\u53ef\u80fd\u6027\u306f\u91cd\u8981\u306a\u5206\u91ce\u3060\u304c\u3001\u73fe\u6642\u70b9\u3067\u6c7a\u5b9a\u6253\u3068\u306a\u308b\u5bfe\u5fdc\u7b56\u306f\u5b58\u5728\u3057\u306a\u3044\u3002<br \/>\n\u672c\u4ef6\u3067\u306f\u5199\u771f\u3092\u64ae\u3063\u305f\u306e\u304c\u81ea\u5206\u81ea\u8eab\u3068\u3044\u3046\u3053\u3068\u3082\u3042\u308a\u3001\u80cc\u666f\u3084\u9262\u304c\u6ce8\u76ee\u70b9\u3068\u306a\u3089\u306a\u3044\u3088\u3046\u6c17\u3092\u4f7f\u3063\u3066\u3044\u308b\u3002\u5177\u4f53\u7684\u306b\u306f\u75c5\u6c17\u306e\u8449\u3068\u5065\u5eb7\u306a\u8449\u305d\u308c\u305e\u308c\u3092\u540c\u3058\u6728\u30fb\u80cc\u666f\u3067\u64ae\u5f71\u3059\u308b\u3001\u69d8\u3005\u306a\u30d1\u30bf\u30fc\u30f3\u3092\u6df7\u305c\u308b\u306a\u3069\u3001\u5909\u306a\u5834\u6240\u306b\u6ce8\u76ee\u3055\u308c\u306a\u3044\u3088\u3046\u306b\u30c7\u30fc\u30bf\u3092\u4f5c\u6210\u3057\u3066\u3044\u308b\u3002<br \/>\n\u30e2\u30c7\u30eb\u306e\u5b66\u7fd2\u6642\u306b\u306f\u5b8c\u5168\u306b\u672a\u77e5\u306e\u30c7\u30fc\u30bf\u3067\u691c\u8a3c\u304c\u3055\u308c\u308b\u3088\u3046\u306b\u3057\u3001\u307e\u305f\u3001\u624b\u52d5\u3067AI\u304c\u9593\u9055\u3044\u3084\u3059\u3044\u30d1\u30bf\u30fc\u30f3\u306e\u89e3\u6790\u30fb\u30c1\u30a7\u30c3\u30af\u3092\u884c\u3063\u3066\u3044\u308b\u3002\u5272\u3068\u4e01\u5be7\u306b\u5224\u5225\u30e2\u30c7\u30eb\u3092\u4f5c\u3063\u3066\u3044\u308b\u304c\u3001\u305d\u308c\u3067\u3082\u3001\u300c\u672a\u77e5\u30c7\u30fc\u30bf\u306b\u5bfe\u3059\u308b\u5224\u5225\u80fd\u529b\u300d\u306f\u300c\u691c\u8a3c\u7d50\u679c\u3068\u3057\u3066\u6570\u5b66\u7684\u306b\u6c42\u3081\u3089\u308c\u305f\u80fd\u529b\u300d\u307b\u3069\u9ad8\u304f\u306f\u306a\u3044\u3093\u3060\u308d\u3046\u306a\u30fc\u3068\u601d\u3063\u3066\u3044\u308b\u3002<br \/>\n<!--more--><\/p>\n<h2>\u305d\u306e\u4ed6<\/h2>\n<p>\u30e2\u30c7\u30eb\u69cb\u7bc9\u7528\u306e\u30c7\u30fc\u30bf\u3092\u81ea\u5206\u3067\u64ae\u5f71\u3057\u305f\u306e\u306f\u826f\u3044\u7d4c\u9a13\u306b\u306a\u3063\u305f\u3002\u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u30dd\u30a4\u30f3\u30c8\u306b\u3042\u3052\u305f\u4e8b\u9805\u306f\u77e5\u8b58\u3068\u3057\u3066\u306f\u77e5\u3063\u3066\u3044\u3066\u3001\u5bfe\u5fdc\u3082\u8272\u3005\u884c\u3063\u3066\u304d\u305f\u3002\u3057\u304b\u3057\u3001\u5b9f\u969b\u306e\u6240\u3069\u306e\u7a0b\u5ea6\u60aa\u3055\u3092\u3057\u3001\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u5de5\u592b\u3067\u3069\u306e\u7a0b\u5ea6\u56de\u907f\u3067\u304d\u308b\u304b\u3092\u8a66\u3059\u6a5f\u4f1a\u306f\u591a\u304f\u306a\u3044\u3002\u81ea\u5206\u3067\u64ae\u5f71\u30fb\u81ea\u5206\u3067\u30e9\u30d9\u30ea\u30f3\u30b0\u30fb\u81ea\u5206\u3067\u30e2\u30c7\u30eb\u69cb\u7bc9\u3068\u4e00\u9023\u306e\u6d41\u308c\u3092\u3084\u3063\u3066\u307f\u308b\u3068\u611f\u899a\u3092\u3064\u304b\u3080\u3053\u3068\u304c\u3067\u304d\u308b\u3002\uff08\u590f\u306e\u64ae\u5f71\u306f\u6691\u304b\u3063\u305f\u304c\u30fb\u30fb\u30fb\u3002\uff09<br \/>\n\u8a18\u4e8b\u3092\u66f8\u304d\u3064\u3064\u81ea\u5206\u3067\u3082\u7591\u554f\u306b\u601d\u3063\u305f\u306e\u3060\u304c\u3001GlobalAveragePooling2D\u3057\u3066\u304b\u3089\u3001\u3055\u3089\u306b\u4f55\u5c64\u304b\u8ffd\u52a0\u3057\u305f\u7406\u7531\u3084\u5909\u66f4\u53ef\u80fd\u5c64\u3092\u591a\u304f\u3068\u3063\u3066\u3044\u308b\u7406\u7531\u304c\u601d\u3044\u51fa\u305b\u306a\u3044\u30fb\u30fb\u30fb\u3002\u7cbe\u5ea6\u5411\u4e0a\u306b\u52b9\u679c\u304c\u3042\u3063\u305f\uff01\u3068\u304b\u3060\u3063\u305f\u3088\u3046\u306a\u6c17\u304c\u3059\u308b\u304c\u3001\u5f53\u6642\u306ejupyter-notebook\u306b\u6b8b\u3063\u3066\u3044\u306a\u304b\u3063\u305f\u70b9\u306f\u53cd\u7701\u3002<br \/>\n*1)\u300cAI\u306f\u30d0\u30e9\u304c\u690d\u308f\u3063\u3066\u3044\u308b\u9262\u306e\u8272\u306b\u6ce8\u76ee\u3057\u3066\u75c5\u6c17\u3060\u3068\u5224\u5225\u3059\u308b\u300d\u3088\u3046\u306a\u72b6\u6cc1\u306f\u826f\u304f\u3042\u308b\u3002\u4f8b\u3048\u3070\u300c\u5e74\u5ea6\u3054\u3068\u306b\u9055\u3046\u9262\u3092\u4f7f\u3046\u516c\u5712\u3067\u300d\u300c\u3042\u308b\u5e74\u306b\u4f7f\u3063\u305f\u571f\u306b\u75c5\u539f\u4f53\u304c\u591a\u304f\u3044\u3066\u300d\u300c\u3042\u308b\u7a2e\u306e\u9262\u306b\u690d\u308f\u3063\u3066\u3044\u308b\u30d0\u30e9\u306e\u767a\u75c5\u7387\u304c\u9ad8\u3044\u300d\u5834\u5408\u3001\u8fd1\u5e74\u306eAI\u306f\u300c\u9262\u306e\u7279\u5fb4\u306f\u75c5\u6c17\u3068\u95a2\u9023\u6027\u304c\u9ad8\u3044\u300d\u3068\u3042\u308b\u610f\u5473\u6b63\u3057\u304f\u95a2\u9023\u3092\u898b\u3064\u3051\u308b\u3002\u5b66\u7fd2\u30c7\u30fc\u30bf\u306b\u5bfe\u3057\u3066\u306f\u6b63\u3057\u3044\u52d5\u304d\u306a\u306e\u3060\u304c\u3001\u3053\u306eAI\u306f\u5b66\u7fd2\u30c7\u30fc\u30bf\u3092\u53d6\u3063\u305f\u516c\u5712\u4ee5\u5916\u3067\u306f\u3046\u307e\u304f\u52d5\u304b\u306a\u3044\uff08\u4ed6\u306e\u516c\u5712\u304c\u524d\u8ff0\u306e\u3088\u3046\u306b\u9262\u3092\u9078\u3093\u3067\u3044\u308b\u308f\u3051\u3067\u306f\u306a\u3044\u306e\u3067\uff09\u3002\u5229\u7528\u7528\u9014\u3092\u8003\u3048\u306a\u304c\u3089\u3001\u59a5\u5f53\u306a\u30c7\u30fc\u30bf\u3092\u9078\u3073\u3001\u59a5\u5f53\u306b\u30e2\u30c7\u30eb\u3092\u4f5c\u308a\u3001\u59a5\u5f53\u306b\u904b\u7528\u3059\u308b\u3053\u3068\u306f\u7c21\u5358\u3067\u306f\u306a\u3044\u3002\u3002\u3002<br \/>\n*1)\u306e\u3055\u3089\u306b\u88dc\u8db3\u3002\u300c\u9262\u306e\u8272\u3068\u571f\u306e\u6e29\u5ea6\u306f\u95a2\u9023\u304c\u3042\u308b\u305f\u3081\u3001\u9262\u3068\u767a\u75c5\u7387\u306b\u95a2\u9023\u304c\u3042\u308b\u306e\u306f\u5f53\u7136\u300d\u300c\u9262\u3058\u3083\u306a\u304f\u3066\u5f71\u3092\u65e5\u5f53\u305f\u308a\u3092\u898b\u3066\u3044\u308b\u306e\u3060\uff01\u300d\u306a\u3069\u3001AI\u304c\u51fa\u3057\u305f\u7d50\u679c\u306e\u89e3\u91c8\u306f\u69d8\u3005\u5206\u304b\u308c\u308b\u3053\u3068\u306f\u591a\u3044\u3002\u30c9\u30e1\u30a4\u30f3\u77e5\u8b58\u306f\u4eca\u3067\u3082\u91cd\u8981\u3067\u3042\u308b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u524d\u56de\u304b\u3089\u5f15\u304d\u7d9a\u304d\u3001\u690d\u7269\u306e\u5199\u771f\u304b\u3089\u75c5\u6c17\u3092\u5224\u5225\u3059\u308b\u30b5\u30a4\u30c8(http:\/\/www.plant-check.jp\/)\u3092\u4f5c\u3063\u305f\u3068\u304d\u306e\u307e\u3068\u3081\u3002 \u30e2\u30c7\u30eb\u69cb\u7bc9\u306e\u6d41\u308c \u524d\u56de\u66f8\u3044\u305f\u3068\u304a\u308a\u3001\u4eca\u56de\u3001(1)\u690d\u7269\u306e\u8449\u304b\u305d\u308c\u4ee5\u5916\u304b\u3092\u5224\u5225\u3059\u308b2\u5024\u5206\u985e\u30e2 &hellip; <a href=\"https:\/\/staka.jp\/wordpress\/?p=102\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;\u30e2\u30c7\u30eb\u69cb\u7bc9\uff08keras\/tensorflow+InceptionV3+Data augmentation\uff09\u7de8\uff082\/3\uff09&#8221; \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2,8,11],"tags":[27,40,69],"class_list":["post-102","post","type-post","status-publish","format-standard","hentry","category-ai","category-python","category-11","tag-deeplearning","tag-keras","tag-tensorflow"],"_links":{"self":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/102","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=102"}],"version-history":[{"count":0,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/102\/revisions"}],"wp:attachment":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=102"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=102"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=102"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}