upload codes

CNN-quad2D/codes_keras_v1/config.py

@@ -0,0 +1,14 @@
+GenWeightPath = 'GenCurvedModel.h5'
+DisWeightPath = 'DisCurvedModel.h5'
+CurvedMatPath = 'CNN-quad2D/data/curved_mat/'
+CurvedAnsPath = 'CNN-quad2D/data/curved_ans/'
+
+net_input_size = 32 #256
+row_size = 10
+col_size = 32
+
+train_amount = 400
+test_amount = 100
+
+Epochs = 10 #6000
+BatchSize = 200 #20

CNN-quad2D/codes_keras_v1/model.py

@@ -0,0 +1,74 @@
+import torch
+import os
+from config import *
+from utils import *
+import numpy as np
+import scipy.io as scio
+import tensorflow as tf
+import tensorflow.keras as keras
+from keras.models import Sequential
+from keras.layers import Dense,Dropout,Activation,Flatten
+from keras.layers import Conv2D,GlobalAveragePooling2D,MaxPooling2D,ZeroPadding2D,BatchNormalization
+from tensorflow.keras.optimizers import Adam
+import matplotlib.pyplot as plt
+
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
+np.set_printoptions(threshold=np.inf)
+
+train_data, train_ans, test_data, test_ans = load_data()
+# train_data = train_data.reshape(800,10,32,1)
+train_data = tf.expand_dims(train_data, axis=-1)
+print(train_data.shape)
+
+model=Sequential()
+# model.add(Dense(20))
+# model.add(Activation("relu"))
+# model.add(Dense(1))
+# model.add(Activation("relu"))
+model.add(Conv2D(filters=1,kernel_size=(10,32),
+                 strides=(1,1),padding='valid',
+                 input_shape=(10,32,1),
+                 activation='relu'))
+
+# train the model using Adam
+print("Training network")
+model.compile(loss='mean_squared_logarithmic_error',
+              optimizer='adam',
+              metrics=['accuracy'])
+H = model.fit(train_data,train_ans,validation_data=(test_data,test_ans),
+              epochs=Epochs,batch_size=BatchSize)
+
+model.save_weights("model.h5")
+print("Save model to disk")
+
+# evaluate the network
+print("Evaluating network")
+loss, accuracy = model.evaluate(train_data,train_ans)
+print("\nLoss: %.2f, Accuracy: %.2f%%" % (loss, accuracy*100))
+
+# test
+print("Test network")
+pred = model.predict(test_data,batch_size=BatchSize)
+pred = pred.reshape((200,))
+print(np.abs(pred-test_ans).shape)
+print(test_ans.shape)
+acc = np.mean(np.abs(pred-test_ans) < 1000)
+print(pred.shape)
+print('Prediction Accuracy: %.2f%%' % (acc*100))
+
+
+# # plot the training loss and accuracy
+# plt.style.use("ggplot")
+# plt.figure()
+# plt.plot(np.arange(Epochs), H.history["loss"], label="train_loss")
+# plt.plot(np.arange(Epochs), H.history["val_loss"], label="val_loss")
+# plt.plot(np.arange(Epochs), H.history["accuracy"], label="train_acc")
+# plt.plot(np.arange(Epochs), H.history["val_accuracy"], label="val_acc")
+# plt.title("Training Loss and Accuracy")
+# plt.xlabel("Epoch #")
+# plt.ylabel("Loss/Accuracy")
+# plt.legend()
+# plt.savefig("LOSS.jpg")
+print("Model summary")
+model.summary()
+

CNN-quad2D/codes_keras_v1/test.py

@@ -0,0 +1,29 @@
+import torch
+import os
+import config
+from utils import *
+import tensorflow.keras as keras
+from keras.models import load_model
+from keras.preprocessing import image
+import matplotlib.pyplot as plt
+
+train_data, train_ans, test_data, test_ans = load_data()
+
+print('Using loaded model to predict...')
+model = load_model('model.h5')
+np.set_printoptions(precision=4)
+
+# test
+print("Test network")
+pred = model.predict(test_data,batch_size=BatchSize)
+pred = pred.reshape((200,))
+print(np.abs(pred-test_ans).shape)
+print(test_ans.shape)
+acc = np.mean(np.abs(pred-test_ans) < 1000)
+print(pred.shape)
+print('Prediction Accuracy: %.2f%%' % (acc*100))
+
+model.save('./weights/model.h5')
+
+print("Model summary")
+model.summary()

CNN-quad2D/codes_keras_v1/utils.py

@@ -0,0 +1,53 @@
+import torch
+import numpy as np
+import scipy.io as scio
+from config import *
+
+def load_data():
+    train_data = np.empty((train_amount*2,row_size,col_size),dtype="double")
+    train_ans = np.empty((train_amount*2,),dtype="double")
+    test_data = np.empty((test_amount*2,row_size,col_size),dtype="double")
+    test_ans = np.empty((test_amount*2,),dtype="double")
+
+    print("Load training data")
+    for i in range(1,400):
+        dir = CurvedMatPath + str(i) + ".mat"
+        data = scio.loadmat(dir)
+        train_data[i] = np.pad(data['mat'],((0,6),(0,20)),'constant',constant_values=(0,0))
+        
+        dir = CurvedAnsPath + str(i) + ".mat"
+        data = scio.loadmat(dir)
+        train_ans[i] = data['res']
+        
+        # negtive area
+        dir = CurvedMatPath + str(i) + "_neg.mat"
+        data = scio.loadmat(dir)
+        train_data[i + train_amount] = np.pad(data['mat'],((0,6),(0,20)),'constant',constant_values=(0,0))
+
+        dir = CurvedAnsPath + str(i) + "_neg.mat"
+        data = scio.loadmat(dir)
+        train_ans[i + train_amount] = data['res']
+
+    print("Load test data")
+    for i in range(401,500):
+        dir = CurvedMatPath + str(i) + ".mat"
+        data = scio.loadmat(dir)
+        test_data[i-400] = np.pad(data['mat'],((0,6),(0,20)),'constant',constant_values=(0,0))
+        
+        dir = CurvedAnsPath + str(i) + ".mat"
+        data = scio.loadmat(dir)
+        test_ans[i-400] = data['res']
+        
+        # negtive area
+        dir = CurvedMatPath + str(i) + "_neg.mat"
+        data = scio.loadmat(dir)
+        test_data[i-400+test_amount] = np.pad(data['mat'],((0,6),(0,20)),'constant',constant_values=(0,0))
+
+        dir = CurvedAnsPath + str(i) + "_neg.mat"
+        data = scio.loadmat(dir)
+        test_ans[i-400+test_amount] = data['res']
+    if np.any(np.isnan(train_data)) or np.any(np.isnan(train_ans)) or np.any(np.isnan(test_data)) or np.any(np.isnan(test_ans)):
+        print("EXIST NAN")
+    
+    return train_data, train_ans, test_data, test_ans
+