In [2]:
display_image_data(config.DATASET_PATH)
Out[2]:
from drig.preprocessors import ImageToArrayPreprocessor, UniformAspectPreprocessor
from drig.datum import ImageDatumLoader
from drig.datum_config import Flowers17Config as config
from drig.config import TinyVGGNetImage, ImageCast
from sklearn.preprocessing import LabelBinarizer, LabelEncoder
from sklearn.model_selection import train_test_split
from drig.networks import TinyVGGNet
from keras.optimizers import SGD
from sklearn.metrics import classification_report
from drig.utils import plot_training_metrics, plot_network, \
display_image_data, display_prediction, display_image, \
visualize_network, plot_confusion_mesh, confusion_mesh, random_image, display_prediction, preprocess_image
display_image_data(config.DATASET_PATH)
config.IMAGE_PREPROCESSING_HEIGHT,config.IMAGE_PREPROCESSING_WIDTH
(64, 64)
uniform_aspect_preproc=UniformAspectPreprocessor(config.IMAGE_PREPROCESSING_HEIGHT,
config.IMAGE_PREPROCESSING_WIDTH)
image_to_array_preproc=ImageToArrayPreprocessor()
preprocessors=[uniform_aspect_preproc, image_to_array_preproc]
data_loader=ImageDatumLoader(preprocessors=preprocessors)
data, labels= data_loader.load(dataset_path=config.DATASET_PATH,
class_index=config.CLASS_INDEX,
input_cast=ImageCast.RGB_64x64)
Loading Images :100% |❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆| 1360 of 1360 Time: 0:00:04 INFO:root:PROCESSED 1360/1360 IMAGES. INFO:root:MEMORY SIZE OCCUPIED: 65.28 MB
classes=data_loader.classes
train_x, test_x, train_y, test_y= train_test_split(data, labels, test_size=config.TEST_SIZE, random_state=42)
label_binarizer=LabelBinarizer()
train_y=label_binarizer.fit_transform(train_y)
test_y=label_binarizer.transform(test_y)
refiner=SGD(learning_rate=config.ALPHA)
net=TinyVGGNet.compose(height=TinyVGGNetImage.HEIGHT,
width=TinyVGGNetImage.WIDTH,
depth=TinyVGGNetImage.DEPTH,
classes=len(classes))
net.compile(loss="categorical_crossentropy", optimizer=refiner, metrics=["accuracy"])
visualize_network(net)
Z=net.fit(train_x, train_y, batch_size=config.BATCH_SIZE, validation_data=(test_x, test_y),
epochs=config.EPOCHS, verbose=1)
Epoch 1/100 16/16 [==============================] - 3s 153ms/step - loss: 3.5782 - accuracy: 0.1684 - val_loss: 2.8340 - val_accuracy: 0.0618 Epoch 2/100 16/16 [==============================] - 2s 132ms/step - loss: 1.8038 - accuracy: 0.4656 - val_loss: 2.9538 - val_accuracy: 0.0794 Epoch 3/100 16/16 [==============================] - 2s 132ms/step - loss: 1.3733 - accuracy: 0.5551 - val_loss: 3.0419 - val_accuracy: 0.1147 Epoch 4/100 16/16 [==============================] - 2s 131ms/step - loss: 1.0513 - accuracy: 0.6391 - val_loss: 3.1680 - val_accuracy: 0.1235 Epoch 5/100 16/16 [==============================] - 2s 132ms/step - loss: 0.8887 - accuracy: 0.7065 - val_loss: 3.2795 - val_accuracy: 0.1206 Epoch 6/100 16/16 [==============================] - 2s 131ms/step - loss: 0.6761 - accuracy: 0.7766 - val_loss: 3.2233 - val_accuracy: 0.1441 Epoch 7/100 16/16 [==============================] - 2s 131ms/step - loss: 0.5001 - accuracy: 0.8400 - val_loss: 3.4253 - val_accuracy: 0.1294 Epoch 8/100 16/16 [==============================] - 2s 131ms/step - loss: 0.4731 - accuracy: 0.8454 - val_loss: 3.4444 - val_accuracy: 0.1471 Epoch 9/100 16/16 [==============================] - 2s 130ms/step - loss: 0.3348 - accuracy: 0.9041 - val_loss: 3.5602 - val_accuracy: 0.1206 Epoch 10/100 16/16 [==============================] - 2s 131ms/step - loss: 0.3098 - accuracy: 0.9030 - val_loss: 3.5612 - val_accuracy: 0.1353 Epoch 11/100 16/16 [==============================] - 2s 131ms/step - loss: 0.2713 - accuracy: 0.9195 - val_loss: 3.4998 - val_accuracy: 0.1912 Epoch 12/100 16/16 [==============================] - 2s 131ms/step - loss: 0.2517 - accuracy: 0.9160 - val_loss: 3.5464 - val_accuracy: 0.1559 Epoch 13/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1904 - accuracy: 0.9628 - val_loss: 3.6593 - val_accuracy: 0.1941 Epoch 14/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1841 - accuracy: 0.9572 - val_loss: 3.5509 - val_accuracy: 0.1971 Epoch 15/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1740 - accuracy: 0.9447 - val_loss: 3.4387 - val_accuracy: 0.2029 Epoch 16/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1678 - accuracy: 0.9531 - val_loss: 3.4438 - val_accuracy: 0.2176 Epoch 17/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1163 - accuracy: 0.9719 - val_loss: 3.6315 - val_accuracy: 0.2029 Epoch 18/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1212 - accuracy: 0.9716 - val_loss: 3.4227 - val_accuracy: 0.2294 Epoch 19/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1531 - accuracy: 0.9565 - val_loss: 3.3993 - val_accuracy: 0.2235 Epoch 20/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0905 - accuracy: 0.9825 - val_loss: 3.4444 - val_accuracy: 0.2265 Epoch 21/100 16/16 [==============================] - 2s 131ms/step - loss: 0.1342 - accuracy: 0.9653 - val_loss: 3.2153 - val_accuracy: 0.2588 Epoch 22/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0958 - accuracy: 0.9779 - val_loss: 2.9839 - val_accuracy: 0.2735 Epoch 23/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0813 - accuracy: 0.9799 - val_loss: 2.9035 - val_accuracy: 0.2882 Epoch 24/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0761 - accuracy: 0.9804 - val_loss: 2.8836 - val_accuracy: 0.3235 Epoch 25/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0699 - accuracy: 0.9897 - val_loss: 2.8337 - val_accuracy: 0.3294 Epoch 26/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0639 - accuracy: 0.9902 - val_loss: 2.5631 - val_accuracy: 0.3588 Epoch 27/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0753 - accuracy: 0.9864 - val_loss: 2.4317 - val_accuracy: 0.3676 Epoch 28/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0670 - accuracy: 0.9888 - val_loss: 2.3505 - val_accuracy: 0.3882 Epoch 29/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0627 - accuracy: 0.9881 - val_loss: 2.2354 - val_accuracy: 0.4206 Epoch 30/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0664 - accuracy: 0.9860 - val_loss: 2.1766 - val_accuracy: 0.4324 Epoch 31/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0454 - accuracy: 0.9918 - val_loss: 2.0792 - val_accuracy: 0.4559 Epoch 32/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0476 - accuracy: 0.9958 - val_loss: 2.0729 - val_accuracy: 0.4676 Epoch 33/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0491 - accuracy: 0.9894 - val_loss: 1.8977 - val_accuracy: 0.4971 Epoch 34/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0412 - accuracy: 0.9931 - val_loss: 1.8971 - val_accuracy: 0.5265 Epoch 35/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0488 - accuracy: 0.9955 - val_loss: 1.8771 - val_accuracy: 0.5265 Epoch 36/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0398 - accuracy: 0.9905 - val_loss: 1.8149 - val_accuracy: 0.5324 Epoch 37/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0450 - accuracy: 0.9887 - val_loss: 1.8163 - val_accuracy: 0.5559 Epoch 38/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0337 - accuracy: 0.9953 - val_loss: 1.8122 - val_accuracy: 0.5676 Epoch 39/100 16/16 [==============================] - 2s 134ms/step - loss: 0.0476 - accuracy: 0.9880 - val_loss: 1.8264 - val_accuracy: 0.5647 Epoch 40/100 16/16 [==============================] - 2s 132ms/step - loss: 0.0349 - accuracy: 0.9966 - val_loss: 1.7556 - val_accuracy: 0.5676 Epoch 41/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0386 - accuracy: 0.9965 - val_loss: 1.7302 - val_accuracy: 0.5824 Epoch 42/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0376 - accuracy: 0.9950 - val_loss: 1.7346 - val_accuracy: 0.5853 Epoch 43/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0268 - accuracy: 0.9994 - val_loss: 1.7470 - val_accuracy: 0.5853 Epoch 44/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0290 - accuracy: 0.9940 - val_loss: 1.7301 - val_accuracy: 0.5941 Epoch 45/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0310 - accuracy: 0.9927 - val_loss: 1.7412 - val_accuracy: 0.5912 Epoch 46/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0399 - accuracy: 0.9883 - val_loss: 1.7815 - val_accuracy: 0.5941 Epoch 47/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0341 - accuracy: 0.9958 - val_loss: 1.8169 - val_accuracy: 0.5853 Epoch 48/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0253 - accuracy: 0.9985 - val_loss: 1.7501 - val_accuracy: 0.5853 Epoch 49/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0233 - accuracy: 0.9983 - val_loss: 1.7523 - val_accuracy: 0.5882 Epoch 50/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0239 - accuracy: 0.9990 - val_loss: 1.7079 - val_accuracy: 0.5941 Epoch 51/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0276 - accuracy: 0.9939 - val_loss: 1.7031 - val_accuracy: 0.6000 Epoch 52/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0280 - accuracy: 0.9935 - val_loss: 1.7144 - val_accuracy: 0.6000 Epoch 53/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0231 - accuracy: 0.9957 - val_loss: 1.7414 - val_accuracy: 0.6000 Epoch 54/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0257 - accuracy: 0.9969 - val_loss: 1.7975 - val_accuracy: 0.5912 Epoch 55/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0257 - accuracy: 0.9978 - val_loss: 1.7821 - val_accuracy: 0.5941 Epoch 56/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0172 - accuracy: 0.9998 - val_loss: 1.7247 - val_accuracy: 0.6059 Epoch 57/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0239 - accuracy: 0.9963 - val_loss: 1.7234 - val_accuracy: 0.5912 Epoch 58/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0182 - accuracy: 0.9971 - val_loss: 1.7216 - val_accuracy: 0.5912 Epoch 59/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0299 - accuracy: 0.9922 - val_loss: 1.7610 - val_accuracy: 0.5853 Epoch 60/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0262 - accuracy: 0.9945 - val_loss: 1.7895 - val_accuracy: 0.5941 Epoch 61/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0189 - accuracy: 0.9968 - val_loss: 1.8039 - val_accuracy: 0.5941 Epoch 62/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0184 - accuracy: 0.9977 - val_loss: 1.8149 - val_accuracy: 0.5971 Epoch 63/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0242 - accuracy: 0.9968 - val_loss: 1.7678 - val_accuracy: 0.6118 Epoch 64/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0257 - accuracy: 0.9965 - val_loss: 1.7592 - val_accuracy: 0.6059 Epoch 65/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0210 - accuracy: 0.9953 - val_loss: 1.7880 - val_accuracy: 0.6118 Epoch 66/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0190 - accuracy: 0.9997 - val_loss: 1.7814 - val_accuracy: 0.6088 Epoch 67/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0190 - accuracy: 0.9952 - val_loss: 1.9370 - val_accuracy: 0.5765 Epoch 68/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0224 - accuracy: 0.9959 - val_loss: 1.8227 - val_accuracy: 0.5941 Epoch 69/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0142 - accuracy: 0.9977 - val_loss: 1.8509 - val_accuracy: 0.5971 Epoch 70/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0158 - accuracy: 0.9968 - val_loss: 1.8380 - val_accuracy: 0.5941 Epoch 71/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0179 - accuracy: 0.9954 - val_loss: 1.8000 - val_accuracy: 0.6059 Epoch 72/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0168 - accuracy: 0.9976 - val_loss: 1.8659 - val_accuracy: 0.5912 Epoch 73/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0119 - accuracy: 0.9989 - val_loss: 1.8296 - val_accuracy: 0.5882 Epoch 74/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0155 - accuracy: 1.0000 - val_loss: 1.7991 - val_accuracy: 0.5971 Epoch 75/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0114 - accuracy: 0.9986 - val_loss: 1.7845 - val_accuracy: 0.6147 Epoch 76/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0153 - accuracy: 0.9970 - val_loss: 1.7756 - val_accuracy: 0.6088 Epoch 77/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0203 - accuracy: 0.9953 - val_loss: 1.7909 - val_accuracy: 0.5971 Epoch 78/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0124 - accuracy: 1.0000 - val_loss: 1.7927 - val_accuracy: 0.6000 Epoch 79/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0134 - accuracy: 0.9993 - val_loss: 1.8271 - val_accuracy: 0.5971 Epoch 80/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0190 - accuracy: 0.9973 - val_loss: 1.8766 - val_accuracy: 0.6000 Epoch 81/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0129 - accuracy: 1.0000 - val_loss: 1.8676 - val_accuracy: 0.5941 Epoch 82/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0141 - accuracy: 1.0000 - val_loss: 1.8547 - val_accuracy: 0.5941 Epoch 83/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0162 - accuracy: 0.9972 - val_loss: 1.9286 - val_accuracy: 0.5941 Epoch 84/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0120 - accuracy: 0.9983 - val_loss: 1.9480 - val_accuracy: 0.5706 Epoch 85/100 16/16 [==============================] - 2s 132ms/step - loss: 0.0110 - accuracy: 0.9995 - val_loss: 1.9044 - val_accuracy: 0.5912 Epoch 86/100 16/16 [==============================] - 2s 130ms/step - loss: 0.0126 - accuracy: 0.9985 - val_loss: 1.8765 - val_accuracy: 0.5971 Epoch 87/100 16/16 [==============================] - 2s 130ms/step - loss: 0.0129 - accuracy: 0.9996 - val_loss: 1.8447 - val_accuracy: 0.6029 Epoch 88/100 16/16 [==============================] - 2s 130ms/step - loss: 0.0147 - accuracy: 0.9992 - val_loss: 1.8641 - val_accuracy: 0.6029 Epoch 89/100 16/16 [==============================] - 2s 130ms/step - loss: 0.0073 - accuracy: 1.0000 - val_loss: 1.8417 - val_accuracy: 0.6088 Epoch 90/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0118 - accuracy: 0.9988 - val_loss: 1.9020 - val_accuracy: 0.5971 Epoch 91/100 16/16 [==============================] - 2s 132ms/step - loss: 0.0131 - accuracy: 0.9986 - val_loss: 1.8860 - val_accuracy: 0.6088 Epoch 92/100 16/16 [==============================] - 2s 131ms/step - loss: 0.0080 - accuracy: 1.0000 - val_loss: 1.8812 - val_accuracy: 0.6000 Epoch 93/100 16/16 [==============================] - 2s 134ms/step - loss: 0.0169 - accuracy: 0.9960 - val_loss: 1.8641 - val_accuracy: 0.5971 Epoch 94/100 16/16 [==============================] - 2s 138ms/step - loss: 0.0071 - accuracy: 0.9995 - val_loss: 1.8544 - val_accuracy: 0.5941 Epoch 95/100 16/16 [==============================] - 2s 135ms/step - loss: 0.0110 - accuracy: 0.9986 - val_loss: 1.8427 - val_accuracy: 0.6029 Epoch 96/100 16/16 [==============================] - 2s 136ms/step - loss: 0.0107 - accuracy: 0.9968 - val_loss: 1.8482 - val_accuracy: 0.6118 Epoch 97/100 16/16 [==============================] - 2s 139ms/step - loss: 0.0102 - accuracy: 1.0000 - val_loss: 1.8301 - val_accuracy: 0.6118 Epoch 98/100 16/16 [==============================] - 2s 133ms/step - loss: 0.0150 - accuracy: 0.9994 - val_loss: 1.8570 - val_accuracy: 0.6147 Epoch 99/100 16/16 [==============================] - 2s 134ms/step - loss: 0.0137 - accuracy: 0.9983 - val_loss: 1.8938 - val_accuracy: 0.6118 Epoch 100/100 16/16 [==============================] - 2s 133ms/step - loss: 0.0097 - accuracy: 0.9998 - val_loss: 1.8872 - val_accuracy: 0.6088
predictions=net.predict(test_x, batch_size=config.BATCH_SIZE)
print(classification_report(test_y.argmax(axis=1),predictions.argmax(axis=1), target_names=classes))
precision recall f1-score support
0 0.29 0.33 0.31 15
1 0.62 0.65 0.63 20
10 0.79 0.70 0.75 27
11 0.40 0.70 0.51 20
12 0.60 0.94 0.73 16
13 0.47 0.39 0.42 18
14 0.82 0.39 0.53 23
15 0.70 0.70 0.70 20
16 0.75 0.63 0.69 19
2 0.71 0.56 0.63 18
3 0.83 0.45 0.59 22
4 0.54 0.75 0.63 20
5 0.88 0.65 0.75 23
6 0.86 0.57 0.69 21
7 0.27 0.35 0.30 23
8 0.52 0.75 0.62 16
9 0.94 0.89 0.92 19
accuracy 0.61 340
macro avg 0.65 0.61 0.61 340
weighted avg 0.66 0.61 0.61 340
plot_training_metrics(Z,epochs=config.EPOCHS, inline=True)
mesh, class_name,class_index=confusion_mesh(test_y.argmax(axis=1),
predictions.argmax(axis=1),
encoded_classes=LabelEncoder().fit_transform(classes),
class_name="16",
classes=classes.tolist())
plot_confusion_mesh(mesh, class_name)
test_image, class_name=random_image(config.DATASET_PATH,class_index=config.CLASS_INDEX)
display_image(image=test_image)
class_name
'16'
preprocessed_test_image=preprocess_image(preprocessors=preprocessors,image=test_image, for_prediction=True)
test_predict=net.predict(preprocessed_test_image)
display_prediction(test_predict.argmax(axis=1),classes, image=test_image,font_scale=0.7 ,text_thickness=1)
