Skip to content

Commit 897b374

Browse files
arvcodearvcode
committed
Commit Message:
Jupyter Notebook for CNN using Keras and TensorFlow2.0. Change log: MNIST Handwritten Number detection. Committed By: Arvindh
1 parent 81d0c72 commit 897b374

File tree

2 files changed

+346
-0
lines changed

2 files changed

+346
-0
lines changed

‎CNN using Keras and TensorFlow2.0_MNIST.ipynb

+346
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,346 @@
1+
{
2+
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"## MNIST Handwritten Numbers detection via CNN using Keras,TensorFlow 2.0"
8+
]
9+
},
10+
{
11+
"cell_type": "code",
12+
"execution_count": null,
13+
"metadata": {},
14+
"outputs": [],
15+
"source": [
16+
"#Import Modules\n",
17+
"import tensorflow as tf\n",
18+
"import cv2 as cv\n",
19+
"import numpy as np\n",
20+
"import matplotlib.pyplot as plt\n",
21+
"import copy\n",
22+
"\n",
23+
"#Dataset.\n",
24+
"from tensorflow.keras.datasets import mnist\n",
25+
"#One-hot encoding.\n",
26+
"from tensorflow.keras.utils import to_categorical\n",
27+
"#Feedforward sequential network with no feedback loop.\n",
28+
"from tensorflow.keras.models import Sequential\n",
29+
"#Layers.\n",
30+
"from tensorflow.keras.layers import Dense, Dropout, Flatten\n",
31+
"from tensorflow.keras.layers import Conv2D, MaxPooling2D\n",
32+
"#Optimizer.\n",
33+
"from tensorflow.keras.optimizers import SGD\n",
34+
"from tensorflow.keras import backend as K\n",
35+
"#Model loader.\n",
36+
"from tensorflow.keras.models import load_model\n",
37+
"\n"
38+
]
39+
},
40+
{
41+
"cell_type": "code",
42+
"execution_count": null,
43+
"metadata": {},
44+
"outputs": [],
45+
"source": [
46+
"#load the MNIST handwritten number dataset.\n",
47+
"(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
48+
"print(x_train.shape)\n",
49+
"print(x_test.shape)\n",
50+
"print(x_train[0].shape)"
51+
]
52+
},
53+
{
54+
"cell_type": "code",
55+
"execution_count": null,
56+
"metadata": {},
57+
"outputs": [],
58+
"source": [
59+
"#Display random samples from training set.\n",
60+
"\n",
61+
"#pyplot subplot nrows,ncols,index->3,3,1\n",
62+
"subplot_num=331\n",
63+
"for index in range(0,5):\n",
64+
" random_num=np.random.randint(0,len(x_train))\n",
65+
" image=x_train[random_num]\n",
66+
" plt.subplot(subplot_num)\n",
67+
" plt.imshow(image,cmap='gray')\n",
68+
" subplot_num=subplot_num+1"
69+
]
70+
},
71+
{
72+
"cell_type": "code",
73+
"execution_count": null,
74+
"metadata": {},
75+
"outputs": [],
76+
"source": [
77+
"#Preprocess the data.\n",
78+
"#Convert the image shape for Keras.\n",
79+
"#Convert labels to one-hot encoded representation.\n",
80+
"#Normalize the data.\n",
81+
"\n",
82+
"#Convert for Keras.\n",
83+
"img_rows=x_train[0].shape[0]\n",
84+
"img_cols=x_train[0].shape[1]\n",
85+
"input_img_shape=(img_rows,img_cols,1)\n",
86+
"\n",
87+
"#Reshape.\n",
88+
"x_train=x_train.reshape(x_train.shape[0],img_rows,img_cols,1)\n",
89+
"x_test=x_test.reshape(x_test.shape[0],img_rows,img_cols,1)\n",
90+
"\n",
91+
"#One Hot encoding.\n",
92+
"print('Before one-hot encoding',y_train.shape[0])\n",
93+
"\n",
94+
"y_train=to_categorical(y_train)\n",
95+
"y_test=to_categorical(y_test)\n",
96+
"\n",
97+
"print(\"After one-hot encoding\",y_train.shape[0],y_train.shape[1])\n",
98+
"num_classes=y_train.shape[1]\n",
99+
"num_pixels=x_train.shape[0]*x_train.shape[1]\n",
100+
"\n",
101+
"#Normalize the data.\n",
102+
"print(x_test.dtype)\n",
103+
"print(x_train.dtype)\n",
104+
"x_train=x_train.astype('float32')\n",
105+
"x_test=x_test.astype('float32')\n",
106+
"\n",
107+
"x_train=x_train/255.0\n",
108+
"x_test=x_test/255.0\n",
109+
"\n"
110+
]
111+
},
112+
{
113+
"cell_type": "code",
114+
"execution_count": null,
115+
"metadata": {},
116+
"outputs": [],
117+
"source": [
118+
"#Build the model.\n",
119+
"\n",
120+
"model=Sequential()\n",
121+
"\n",
122+
"#First layer.\n",
123+
"#padding=0, stride=1\n",
124+
"#(nxn)*(f*f) =[((n+2p-f)/s) +1, ((n+2p-f)/s) +1]\n",
125+
"# (28x28x1)*(3x3x32)= (26x26x32)\n",
126+
"model.add(Conv2D(32, kernel_size=(3,3), activation='relu',input_shape=input_img_shape))\n",
127+
"\n",
128+
"#Second layer.(24x24x64)\n",
129+
"model.add(Conv2D(64,(3,3),activation='relu'))\n",
130+
"\n",
131+
"#Max-Pooling.(12x12x64)\n",
132+
"model.add(MaxPooling2D(pool_size=(2,2)))\n",
133+
"\n",
134+
"#Dropout\n",
135+
"model.add(Dropout(0.25))\n",
136+
"\n",
137+
"#Flatten 9216x1\n",
138+
"model.add(Flatten())\n",
139+
"\n",
140+
"#Dense Layer (1x128)\n",
141+
"model.add(Dense(128,activation='relu'))\n",
142+
"\n",
143+
"#Dropout.\n",
144+
"model.add(Dropout(0.5))\n",
145+
"\n",
146+
"#Final layer \n",
147+
"model.add(Dense(num_classes,activation='softmax'))\n",
148+
"\n",
149+
"#Compile the model.(SGD, LR=0.01, loss=CCE)\n",
150+
"model.compile(loss='categorical_crossentropy', optimizer=SGD(0.01), metrics=['accuracy'])\n",
151+
"\n",
152+
"print(model.summary())"
153+
]
154+
},
155+
{
156+
"cell_type": "code",
157+
"execution_count": null,
158+
"metadata": {},
159+
"outputs": [],
160+
"source": [
161+
"#Training the model\n",
162+
"batch=5\n",
163+
"epochs=10\n",
164+
"\n",
165+
"history=model.fit(x_train,y_train,batch_size=batch,epochs=epochs,verbose=1,validation_data=(x_test,y_test))\n"
166+
]
167+
},
168+
{
169+
"cell_type": "code",
170+
"execution_count": null,
171+
"metadata": {},
172+
"outputs": [],
173+
"source": [
174+
"#Evaluate the model\n",
175+
"score=model.evaluate(x_test,y_test,verbose=1)\n",
176+
"\n",
177+
"print('Test accuracy',score[1])\n",
178+
"print('Test loss',score[0])"
179+
]
180+
},
181+
{
182+
"cell_type": "code",
183+
"execution_count": null,
184+
"metadata": {},
185+
"outputs": [],
186+
"source": [
187+
"#Plot loss.\n",
188+
"\n",
189+
"history_dict=history.history\n",
190+
"\n",
191+
"loss_values=history_dict['loss']\n",
192+
"#validation loss\n",
193+
"val_loss=history_dict['val_loss']\n",
194+
"\n",
195+
"epochs=range(1,len(loss_values)+1)\n",
196+
"\n",
197+
"line1=plt.plot(epochs,val_loss,label='Validation loss')\n",
198+
"line2=plt.plot(epochs,loss_values,label='Training loss')\n",
199+
"plt.setp(line1,linewidth=2.0,marker='+', markersize=10.0)\n",
200+
"plt.setp(line2,linewidth=2.0,marker='3',markersize=10.0)\n",
201+
"plt.xlabel('Epochs')\n",
202+
"plt.ylabel('Loss')\n",
203+
"plt.grid(True)\n",
204+
"plt.legend()\n",
205+
"plt.show()"
206+
]
207+
},
208+
{
209+
"cell_type": "code",
210+
"execution_count": null,
211+
"metadata": {},
212+
"outputs": [],
213+
"source": [
214+
"#Plot accuracy\n",
215+
"history_dict = history.history\n",
216+
"\n",
217+
"acc_values = history_dict['accuracy']\n",
218+
"val_acc_values = history_dict['val_accuracy']\n",
219+
"epochs = range(1, len(loss_values) + 1)\n",
220+
"\n",
221+
"line1 = plt.plot(epochs, val_acc_values, label='Validation/Test Accuracy')\n",
222+
"line2 = plt.plot(epochs, acc_values, label='Training Accuracy')\n",
223+
"plt.setp(line1, linewidth=2.0, marker = '+', markersize=10.0)\n",
224+
"plt.setp(line2, linewidth=2.0, marker = '4', markersize=10.0)\n",
225+
"plt.xlabel('Epochs') \n",
226+
"plt.ylabel('Accuracy')\n",
227+
"plt.grid(True)\n",
228+
"plt.legend()\n",
229+
"plt.show()"
230+
]
231+
},
232+
{
233+
"cell_type": "code",
234+
"execution_count": null,
235+
"metadata": {},
236+
"outputs": [],
237+
"source": [
238+
"#Save the model.\n",
239+
"model.save('mnist_number_trained_model.h5')\n",
240+
"\n"
241+
]
242+
},
243+
{
244+
"cell_type": "code",
245+
"execution_count": null,
246+
"metadata": {},
247+
"outputs": [],
248+
"source": [
249+
"#load the saved model\n",
250+
"classifier=load_model('mnist_number_trained_model.h5')"
251+
]
252+
},
253+
{
254+
"cell_type": "code",
255+
"execution_count": null,
256+
"metadata": {},
257+
"outputs": [],
258+
"source": [
259+
"#Test on test data.\n",
260+
"test_image=x_test[3]\n",
261+
"print(test_image.shape)\n",
262+
"#Reshape for Keras.\n",
263+
"test_image=test_image.reshape(1,28,28,1)\n",
264+
"\n",
265+
"#res=classifier.predict_classes(test_image,1,verbose=0)\n",
266+
"#res=str(res[0])\n",
267+
"#print(res)\n",
268+
"\n",
269+
"res=classifier.predict(test_image)\n",
270+
"print(res)\n",
271+
"res=np.argmax(res)\n",
272+
"print(res)\n",
273+
"\n",
274+
"test_image=test_image.reshape(28,28)\n",
275+
"plt.imshow(test_image,cmap='gray')\n",
276+
"print('Prediction',res)"
277+
]
278+
},
279+
{
280+
"cell_type": "code",
281+
"execution_count": null,
282+
"metadata": {},
283+
"outputs": [],
284+
"source": [
285+
"cap = cv.VideoCapture(0)\n",
286+
"\n",
287+
"while True:\n",
288+
"\n",
289+
" ret, frame = cap.read()\n",
290+
" \n",
291+
" #define region of interest\n",
292+
" roi = frame[100:400, 320:620]\n",
293+
" cv.imshow('roi', roi)\n",
294+
" roi = cv.cvtColor(roi, cv.COLOR_BGR2GRAY)\n",
295+
" roi = cv.resize(roi, (28, 28), interpolation = cv.INTER_AREA)\n",
296+
" \n",
297+
" cv.imshow('roi sacled and gray', roi)\n",
298+
" copy = frame.copy()\n",
299+
" cv.rectangle(copy, (320, 100), (620, 400), (255,0,0), 5)\n",
300+
" \n",
301+
" roi = roi.reshape(1,28,28,1) \n",
302+
"\n",
303+
" result = classifier.predict(roi)\n",
304+
" result =np.argmax(result)\n",
305+
" #print(result)\n",
306+
" result=str(result)[0]\n",
307+
" cv.putText(copy, result, (300 , 100), cv.FONT_HERSHEY_COMPLEX, 2, (0, 255, 0), 2)\n",
308+
" cv.imshow('frame', copy) \n",
309+
" \n",
310+
" if cv.waitKey(1) == 13: #13 is the Enter Key\n",
311+
" break\n",
312+
" \n",
313+
"cap.release()\n",
314+
"cv.destroyAllWindows() "
315+
]
316+
},
317+
{
318+
"cell_type": "code",
319+
"execution_count": null,
320+
"metadata": {},
321+
"outputs": [],
322+
"source": []
323+
}
324+
],
325+
"metadata": {
326+
"kernelspec": {
327+
"display_name": "Python 3",
328+
"language": "python",
329+
"name": "python3"
330+
},
331+
"language_info": {
332+
"codemirror_mode": {
333+
"name": "ipython",
334+
"version": 3
335+
},
336+
"file_extension": ".py",
337+
"mimetype": "text/x-python",
338+
"name": "python",
339+
"nbconvert_exporter": "python",
340+
"pygments_lexer": "ipython3",
341+
"version": "3.6.12"
342+
}
343+
},
344+
"nbformat": 4,
345+
"nbformat_minor": 4
346+
}

‎models/mnist_number_trained_model.h5

4.6 MB
Binary file not shown.

0 commit comments

Comments
 (0)