def evaluate_valid(data_op=1, op=1, reduction=1, rate=0.2):
"""
评估函数，在三个数据集上进行评估
:param op: 1-all 2-part
:param data_op: 1-CK 2-Fer 3-Jaffe
:return:
"""
import preprocess
from tqdm import tqdm
from data import CK, Fer2013, Jaffe
filters = Gabor().build_filters()
if data_op == 1:
_, x, y = CK().gen_train_no()
if data_op == 2:
_, x, y = Fer2013().gen_train_no()
if data_op == 3:
_, x, y = Jaffe().gen_train_no()
train = []
if op == 1:
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, reduction)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
if data_op != 2:
# 需要划分
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(train, train, random_state=2019, test_size=rate)
Classifier().SVM(x_train, x_test)
test1 = []
test2 = []
if data_op == 2:
_, x, y = Fer2013().gen_valid_no(1)
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, reduction)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test1.append(res)
_, x, y = Fer2013().gen_valid_no(2)
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, reduction)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test2.append(res)
test1 = np.array(test1)
test2 = np.array(test2)
print("Public")
Classifier().SVM(train, test1)
print("Pirvate")
Classifier().SVM(train, test2)
def evaluate1_lbp():
filters = Gabor().build_filters()
from tqdm import tqdm
from data import CK, Fer2013, Jaffe
_, x, y = Fer2013().gen_train_no()
train = []
for i in tqdm(np.arange(0, x.shape[0], 1)):
res = LBP().get_lbp(x[i])
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
test = []
_, x, y = Jaffe().gen_train_no()
for i in tqdm(np.arange(0, x.shape[0], 1)):
res = LBP().get_lbp(x[i])
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test.append(res)
test = np.array(train)
Classifier().SVM(train, test)
test = []
_, x, y = CK().gen_train_no()
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preliminary_treatment.gray_norm(x[i])
x[i] = preliminary_treatment.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, 6)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test.append(res)
test = np.array(train)
Classifier().SVM(train, test)
def evaluate_test():
"""
在未训练的数据集上进行测试
:return:
"""
filters = Gabor().build_filters()
from tqdm import tqdm
from data import CK, Fer2013, Jaffe
_, x, y = Fer2013().gen_train_no()
train = []
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, 6)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
test = []
_, x, y = Jaffe().gen_train_no()
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, 6)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test.append(res)
test = np.array(train)
Classifier().SVM(train, test)
test = []
_, x, y = CK().gen_train_no()
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preprocess.gray_norm(x[i])
x[i] = preprocess.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, 6)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test.append(res)
test = np.array(train)
Classifier().SVM(train, test)
from visualize import plot_loss, plot_acc
parser = argparse.ArgumentParser()
parser.add_argument("--dataset",
type=str,
default="fer2013",
help="dataset to train, fer2013 or jaffe or ck+")
parser.add_argument("--epochs", type=int, default=200)
parser.add_argument("--batch_size", type=int, default=32)
parser.add_argument("--plot_history", type=bool, default=True)
opt = parser.parse_args()
his = None
print(opt)
if opt.dataset == "fer2013":
expressions, x_train, y_train = Fer2013().gen_train()
_, x_valid, y_valid = Fer2013().gen_valid()
_, x_test, y_test = Fer2013().gen_test()
# target编码
y_train = to_categorical(y_train).reshape(y_train.shape[0], -1)
y_valid = to_categorical(y_valid).reshape(y_valid.shape[0], -1)
# 为了统一几个数据集，必须增加一列为0的
y_train = np.hstack((y_train, np.zeros((y_train.shape[0], 1))))
y_valid = np.hstack((y_valid, np.zeros((y_valid.shape[0], 1))))
print(
"load fer2013 dataset successfully, it has {} train images and {} valid iamges"
.format(y_train.shape[0], y_valid.shape[0]))
model = CNN3(input_shape=(48, 48, 1), n_classes=8)
sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd,
def evaluate_lbp(data_op=1, op=1, reduction=1, rate=0.2):
filters = Gabor().build_filters()
from tqdm import tqdm
from data import CK, Fer2013, Jaffe
_, x, y = CK().gen_train_no()
if data_op == 2:
_, x, y = Fer2013().gen_train_no()
if data_op == 3:
_, x, y = Jaffe().gen_train_no()
train = []
if op == 1:
for i in tqdm(np.arange(0, x.shape[0], 1)):
x[i] = preliminary_treatment.gray_norm(x[i])
x[i] = preliminary_treatment.adaptive_histogram_equalization(x[i])
res = Gabor().getGabor(x[i], filters, False, reduction)
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
if op == 2:
for i in tqdm(np.arange(0, x.shape[0], 1)):
img, dets, shape_list, img_list, pt_post_list = preliminary_treatment.deal(x[i])
res1 = Gabor().getGabor(img, filters, 0, 1)
res1 = np.array(res1)
res = []
if len(shape_list) == 0:
continue
for _, pt in enumerate(shape_list[0].parts()):
px, py = min(max(pt.x, 0), 47), min(max(pt.y, 0), 47)
im = res1[0]
cv2.circle(im, (px, py), 2, (255, 0, 0), 1)
res.append(res1[:, px, py])
res = np.array(res)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
if op == 3:
for i in tqdm(np.arange(0, x.shape[0], 1)):
res = LBP().get_lbp(x[i])
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
train.append(res)
train = np.array(train)
if data_op != 2:
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(train, train, random_state=2019, test_size=rate)
Classifier().SVM(x_train, x_test)
test1 = []
test2 = []
if data_op == 2:
_, x, y = Fer2013().gen_valid_no(1)
for i in tqdm(np.arange(0, x.shape[0], 1)):
res = LBP().get_lbp(x[i])
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test1.append(res)
_, x, y = Fer2013().gen_valid_no(2)
for i in tqdm(np.arange(0, x.shape[0], 1)):
res = LBP().get_lbp(x[i])
res = np.array(res).reshape(-1)
res = np.append(res, y[i])
test2.append(res)
test1 = np.array(test1)
test2 = np.array(test2)
print("Public")
Classifier().SVM(train, test1)
print("Pirvate")
Classifier().SVM(train, test2)