AMLS-assignment/main.py at main · zhonghhao-kuang/AMLS-assignment · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
# add file path of each task to system path
import sys
sys.path += ['./A1/', './A2', './B1', './B2']
from models_a1 import *
from preprocess_a1 import get_a1_data, get_a1_data_cnn
from models_a2 import *
from preprocess_a2 import get_a2_data
from models_b1 import *
from preprocess_b1 import get_b1_data
from models_b2 import *
from preprocess_b2 import get_b2_data, add_class_6_to_label

import warnings
warnings.filterwarnings("ignore")

# define image, label and model path
img_path_A = './Datasets/celeba/img/'
label_path_A = './Datasets/celeba/labels.csv'
test_img_path_A = './Datasets/celeba_test/img/'
test_label_path_A = './Datasets/celeba_test/labels.csv'
a1_saved_model_path = './A1/A1_transfer_test_augment.h5'
a2_saved_model_path = './A2/A2_smiling_72_face_orig_normalizaed.h5'

img_path_B = './Datasets/cartoon_set/img/'
label_path_B = './Datasets/cartoon_set/labels.csv'
test_img_path_B = './Datasets/cartoon_set_test/img/'
test_label_path_B = './Datasets/cartoon_set_test/labels.csv'
b1_saved_model_path = './B1/B1_face_shape_ROI_cnn.h5'
b2_saved_model_path = './B2/B2_eye_color_ROI_cnn.h5'
b2_saved_6_classes_model_path = './B2/B2_eye_color_ROI_cnn_class6.h5'


# ======================================================================================================================


# Task A1
# load data for KNN and SVM, this takes some time
feature_train, feature_test, feature_train_label, feature_test_label = get_a1_data(img_path_A, label_path_A)

# accuracy of knn and svm on split test set
a1_knn_acc, knn_clf_a1 = knn_model_a1(feature_train, feature_test, feature_train_label, feature_test_label, k=4)
a1_svm_acc, svm_clf_a1 = svm_model_a1(feature_train, feature_test, feature_train_label, feature_test_label)

# uncomment below two lines to see grid search process for knn and svm, it may take several minutes to finish
# knn_scores_a1 = grid_search_for_knn_a1(feature_train, feature_test, feature_train_label, feature_test_label, search_num=30)
# best_pred_a1 = grid_search_for_svm_a1(feature_train, feature_test, feature_train_label, cv_folds=5)

# load data for CNN
x_train, x_test, x_valid, y_train, y_test, y_valid = get_a1_data_cnn(img_path_A, label_path_A)

# uncomment this line to train cnn model, checkpointer is commented out in case the pretrained model get covered
# hist_a1 = train_cnn_model_a1(x_train, y_train, x_valid, y_valid)

# accuracy of cnn on split test set
a1_cnn_acc = evaluate_cnn_model_a1(a1_saved_model_path, x_test, y_test)

# acc of three methods on additional test set
acc_of_knn_a1, acc_of_svm_a1, acc_of_cnn_a1 = acc_on_additional_test_set_a1(test_img_path_A, test_label_path_A, a1_saved_model_path, knn_clf_a1, svm_clf_a1)


# ======================================================================================================================


# Task A2
# load data for KNN and SVM
smile_train, smile_test, smile_train_label, smile_test_label = get_a2_data(img_path_A, label_path_A)

# accuracy of knn and svm on split test set
a2_knn_acc, knn_clf_a2 = knn_model_a2(smile_train, smile_test, smile_train_label, smile_test_label, k=11)
a2_svm_acc, svm_clf_a2 = svm_model_a2(smile_train, smile_test, smile_train_label, smile_test_label)
a2_acc_dnn = evaluate_dnn_model_a2(a2_saved_model_path, smile_test, smile_test_label)

# uncomment following two lines to see grid search process for knn and svm
# knn_scores_a2 = grid_search_for_knn_a2(smile_train, smile_test, smile_train_label, smile_test_label, search_num=50)
# best_pred_a2 = grid_search_for_svm_a2(smile_train, smile_test, smile_train_label, cv_folds=5)

# uncomment this line to train cnn model, checkpointer is commented out in case the pretrained model get covered
# hist_a2 = train_dnn_model_a2(smile_train, smile_train_label)

# acc of three methods on additional test set
acc_of_knn_a2, acc_of_svm_a2, acc_of_dnn_a2 = \
    acc_on_additional_test_set_a2(test_img_path_A, test_label_path_A, a2_saved_model_path, knn_clf_a2, svm_clf_a2)


# ======================================================================================================================


# Task B1
# load normalised data
shape_train, shape_test, shape_train_label, shape_test_label = get_b1_data(img_path_B, label_path_B)

# uncomment this line to train the model
# face_shape_history = train_cnn_model_b1(shape_train, shape_train_label)

b1_acc = evaluate_cnn_model_b1(b1_saved_model_path, shape_test, shape_test_label)

# acc on additional test set
acc_of_cnn_b1 = acc_on_additional_test_set_b1(test_img_path_B, test_label_path_B, b1_saved_model_path)


# ======================================================================================================================


# Task B2
# load normalised data
eye_train, eye_test, eye_train_label, eye_test_label = get_b2_data(img_path_B, label_path_B)

# uncomment this line to train the 5d-cnn model
# eye_color_history = train_cnn_model_b2(eye_train, eye_train_label)

# acc of 5d-cnn model on test set
b2_acc = evaluate_cnn_model_b2(b2_saved_model_path, eye_test, eye_test_label)
# acc of 5d-cnn model on additional test set
acc_of_cnn_b2 = acc_on_additional_test_set_b2(test_img_path_B, test_label_path_B, b2_saved_model_path)

# add sunglasses as a new label
eye_color_roi, eye_color_6d_label = add_class_6_to_label(b2_saved_model_path, img_path_B, label_path_B)

# uncomment this line to train the 6d-cnn model
# eye_color_6d_hist = train_cnn_model_b2_6_classes(eye_color_roi, eye_color_6d_label)

# acc of 6d-cnn model on test set
acc_original = evaluate_6d_cnn_model_b2(b2_saved_6_classes_model_path, eye_color_roi, eye_color_6d_label)
# acc of 6d-cnn model on additional test set
acc_test = evaluate_6d_cnn_model_test(b2_saved_model_path, b2_saved_6_classes_model_path,
                                          test_img_path_B, test_label_path_B)


# ======================================================================================================================

print('TA1(test_set)       :KNN:{}, SVM:{}, CNN:{};\nTA1(additional_set) :KNN:{}, SVM:{}, CNN:{};\n\n'
      'TA2(test_set)       :KNN:{}, SVM:{}, DNN:{};\nTA2(additional_set) :KNN:{},  SVM:{}, DNN:{};\n\n'
      'TA3(test_set)       :CNN:{};\nTA3(additional_set) :CNN:{};\n\n'
      'TA4(test_set)       :5d-CNN:{},  6d-CNN:{};\nTA4(additional_set) :5d-CNN:{},  6d-CNN:{};\n'
      .format(a1_knn_acc, a1_svm_acc, a1_cnn_acc,
              acc_of_knn_a1, acc_of_svm_a1, acc_of_cnn_a1,
              a2_knn_acc, a2_svm_acc, a2_acc_dnn,
              acc_of_knn_a2, acc_of_svm_a2, acc_of_dnn_a2,
              b1_acc, acc_of_cnn_b1,
              b2_acc, acc_original, acc_of_cnn_b2, acc_test
             ))