reshapeに関してのエラー

参考サイトを参考に実装しようと考えています。
https://qiita.com/uni-3/items/a62daa5a03a02f5fa46d

python
1## train.py
2import os
3import glob
4import numpy as np
5import keras
6
7from model import SegNet
8
9import dataset
10
11input_shape = (360, 480, 3)
12classes = 12
13epochs = 100
14batch_size = 1
15log_filepath='./logs/'
16
17data_shape = 360*480
18
19class_weighting = [0.2595, 0.1826, 4.5640, 0.1417, 0.5051, 0.3826, 9.6446, 1.8418, 6.6823, 6.2478, 3.0, 7.3614]
20
21## set gpu usage
22import tensorflow as tf
23config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True, per_process_gpu_memory_fraction = 0.8))
24session = tf.Session(config=config)
25keras.backend.tensorflow_backend.set_session(session)
26
27
28def main():
29    print("loading data...")
30    ds = dataset.Dataset(classes=classes,
31                         train_file="CamVid/train.txt",
32                         test_file="CamVid/test.txt")
33    train_X, train_y = ds.load_data(root_path="CamVid",
34                                    mode='train') # need to implement, y shape is (None, 360, 480, classes)
35
36    train_X = ds.preprocess_inputs(train_X)
37    train_Y = ds.reshape_labels(train_y)
38    print("input data shape...", train_X.shape)
39    print("input label shape...", train_Y.shape)
40
41    test_X, test_y = ds.load_data(root_path="CamVid",
42                                  mode='test') # need to implement, y shape is (None, 360, 480, classes)
43    test_X = ds.preprocess_inputs(test_X)
44    test_Y = ds.reshape_labels(test_y)
45
46    tb_cb = keras.callbacks.TensorBoard(log_dir=log_filepath, histogram_freq=1, write_graph=True, write_images=True)
47    print("creating model...")
48    model = SegNet(input_shape=input_shape, classes=classes)
49    model.compile(loss="categorical_crossentropy", optimizer='adadelta', metrics=["accuracy"])
50
51    model.fit(train_X, train_Y, batch_size=batch_size, epochs=epochs,
52              verbose=1, class_weight=class_weighting , validation_data=(test_X, test_Y), shuffle=True
53              , callbacks=[tb_cb])
54
55    model.save('seg.h5')
56
57if __name__ == '__main__':
58    main()

python
1## dataset.py
2import cv2
3import numpy as np
4from pathlib import Path
5from keras.applications import imagenet_utils
6
7
8class Dataset:
9    def __init__(self, classes=12, train_file='train.txt', test_file='test.txt'):
10        self.train_file = train_file
11        self.test_file = test_file
12        self.data_shape = 360*480
13        self.classes = classes
14
15    def normalized(self, rgb):
16        #return rgb/255.0
17        norm=np.zeros((rgb.shape[0], rgb.shape[1], 3),np.float32)
18
19        b=rgb[:,:,0]
20        g=rgb[:,:,1]
21        r=rgb[:,:,2]
22
23        norm[:,:,0]=cv2.equalizeHist(b)
24        norm[:,:,1]=cv2.equalizeHist(g)
25        norm[:,:,2]=cv2.equalizeHist(r)
26
27        return norm
28
29    def one_hot_it(self, labels):
30        x = np.zeros([360,480,12])
31        for i in range(360):
32            for j in range(480):
33                x[i,j,labels[i][j]] = 1
34        return x
35
36    def load_data(self, root_path, mode='train'):
37        data = []
38        label = []
39        root_path = Path(root_path)
40        if (mode == 'train'):
41            filename = self.train_file
42        else:
43            filename = self.test_file
44
45        with open(filename) as f:
46            txt = f.readlines()
47            for line in txt:
48                train_path, label_path = line.split(' ')
49                train_path = root_path.joinpath(train_path[15:])
50                label_path = root_path.joinpath(label_path[15:].strip())
51                data.append(self.normalized(cv2.imread(str(train_path))))
52                label.append(self.one_hot_it(cv2.imread(str(label_path)))[:,:,0])
53                print('.', end='')
54        #print("train data file", os.getcwd() + txt[i][0][7:])
55        #print("label data raw", cv2.imread(os.getcwd() + '/CamVid/trainannot/0001TP_006690.png'))
56        return np.array(data), np.array(label)
57
58
59    def preprocess_inputs(self, X):
60    ### @ https://github.com/fchollet/keras/blob/master/keras/applications/imagenet_utils.py
61        """Preprocesses a tensor encoding a batch of images.
62        # Arguments
63            x: input Numpy tensor, 4D.
64            data_format: data format of the image tensor.
65            mode: One of "caffe", "tf".
66                - caffe: will convert the images from RGB to BGR,
67                    then will zero-center each color channel with
68                    respect to the ImageNet dataset,
69                    without scaling.
70                - tf: will scale pixels between -1 and 1,
71                    sample-wise.
72        # Returns
73            Preprocessed tensor.
74        """
75        return imagenet_utils.preprocess_input(X)
76
77    def reshape_labels(self, y):
78        return np.reshape(y, (len(y), self.data_shape, self.classes))

train.pyを実行後
loading data...
...............................................................................................................................................................................................................................................................................................................................................................................Traceback (most recent call last):
File "train.py", line 58, in <module>
main()
File "train.py", line 37, in main
train_Y = ds.reshape_labels(train_y)
File "/home/ooo/×××/SegNet/dataset.py", line 78, in reshape_labels
return np.reshape(y, (len(y), self.data_shape, self.classes))
File "/home/ooo/anaconda3/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 257, in reshape
return _wrapfunc(a, '0. reshape', newshape, order=order)
File "/home/ooo/anaconda3/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 52, in _wrapfunc
return getattr(obj, method)(*args, **kwds)
ValueError: cannot reshape array of size 63417600 into shape (367,172800,12)

というエラーがでてしまいました。
ValueErrorとして
367*172800=63417600
reshapeしようとする画像データのピクセル数が不足している訳ではなく同じ値なのですがエラーを吐いてしまいます。
参考サイトとして
https://teratail.com/questions/97039
解決策を教えていただけると幸いです。

行動規範の内容に同意します

回答1件

ベストアンサー

修正しました。
動作確認したいのですがGPUのメモリが足りなくなりエラーでこけ、学習が最後まで終わらないため動作の保証はできません。
ちなみに動作確認した環境は
CPU: Ryzen2600X
GPU: GTX1060(6G)
メモリ: 16GB
OS: Widnows10

Python: 3.6.5
Keras: 2.2.0
tensorflow-gpu: 1.8.0
...

python
1# dataset.py
2import cv2
3import numpy as np
4from pathlib import Path
5from keras.applications import imagenet_utils
6
7
8class Dataset:
9    def __init__(self, classes=12, train_file='train.txt', test_file='test.txt'):
10        self.train_file = train_file
11        self.test_file = test_file
12        self.data_shape = 360*480
13        self.classes = classes
14
15    def normalized(self, rgb):
16        #return rgb/255.0
17        norm=np.zeros((rgb.shape[0], rgb.shape[1], 3),np.float32)
18
19        b=rgb[:,:,0]
20        g=rgb[:,:,1]
21        r=rgb[:,:,2]
22
23        norm[:,:,0]=cv2.equalizeHist(b)
24        norm[:,:,1]=cv2.equalizeHist(g)
25        norm[:,:,2]=cv2.equalizeHist(r)
26
27        return norm
28
29    def one_hot_it(self, labels):
30        x = np.zeros([360,480,12])
31        for i in range(360):
32            for j in range(480):
33                x[i,j,labels[i][j]] = 1
34        return x
35
36    def load_data(self, root_path, mode='train'):
37        data = []
38        label = []
39        root_path = Path(root_path)
40        if (mode == 'train'):
41            filename = self.train_file
42        else:
43            filename = self.test_file
44
45        with open(filename) as f:
46            txt = f.readlines()
47            for line in txt:
48                train_path, label_path = line.split(' ')
49                train_path = root_path.joinpath(train_path[15:])
50                label_path = root_path.joinpath(label_path[15:].strip())
51                data.append(self.normalized(cv2.imread(str(train_path))))
52                # 修正部分
53                label.append(self.one_hot_it(cv2.imread(str(label_path))))
54                print('.', end='')
55        #print("train data file", os.getcwd() + txt[i][0][7:])
56        #print("label data raw", cv2.imread(os.getcwd() + '/CamVid/trainannot/0001TP_006690.png'))
57        return np.array(data), np.array(label)
58
59
60    def preprocess_inputs(self, X):
61    ### @ https://github.com/fchollet/keras/blob/master/keras/applications/imagenet_utils.py
62        """Preprocesses a tensor encoding a batch of images.
63        # Arguments
64            x: input Numpy tensor, 4D.
65            data_format: data format of the image tensor.
66            mode: One of "caffe", "tf".
67                - caffe: will convert the images from RGB to BGR,
68                    then will zero-center each color channel with
69                    respect to the ImageNet dataset,
70                    without scaling.
71                - tf: will scale pixels between -1 and 1,
72                    sample-wise.
73        # Returns
74            Preprocessed tensor.
75        """
76        return imagenet_utils.preprocess_input(X)
77
78    def reshape_labels(self, y):
79        return np.reshape(y, (len(y), self.data_shape, self.classes))