add python test directory

@@ -13,9 +13,15 @@ python 코드에서 가중치를 pickle 파일에 저장하는데 c++에서는 pickle 파일을 읽는
 pickletools라는 라이브러리가 있지만 에러가 많고, 자바, 파이썬, C++ 여러 언어를 지원해 무겁기 때문에 속도를 올리기 위한 컨버팅 작업에는 적절하지 않음
 그래서, 파라미터를 저장하는 부분을 csv로 바꿔  C++에서 그 파일로 읽기위해 params.pkl을 params.csv로 바꾸는 코드를 추가함
 
+
 5.  params.pkl->params.txt
 입력처리할 때 csv파일로 읽으면 속도가 느림
 이 또한 속도를 올리기 위한 컨버팅 작업의 목적에 맞지 않기 때문에 params.pkl 파일을 csv 파일이 아닌 txt 파일로 바꿈
 
+
 6. python test 코드 추가
 test하는 부분만 골라내기 위해 python test 코드를 추가(test.py), simple_convnet 내용 추가
+
+
+7. python test 폴더 추가
+python test 폴더에는 test에 필요하지 않은 train 부분을 삭제함 
\ No newline at end of file

+# coding: utf-8
+try:
+    import urllib.request
+except ImportError:
+    raise ImportError('You should use Python 3.x')
+import os.path
+import gzip
+import pickle
+import os
+import numpy as np
+
+
+key_file = {
+    'train':'cifar10-train.gz',
+    'test':'cifar10-test.gz'
+}
+
+dataset_dir = os.path.dirname(os.path.abspath('/Users/HyeonJun/Desktop/simple_convnet/dataset'))
+save_file = dataset_dir + "/cifar10.pkl"
+
+train_num = 50000
+test_num = 10000
+img_dim = (3, 32, 32)
+img_size = 3072
+
+def _load_label(file_name):
+    file_path = dataset_dir + "/" + file_name
+
+    print("Converting " + file_name + " to NumPy Array ...")
+    with gzip.open(file_path, 'rb') as f:
+            labels = np.frombuffer(f.read(), np.uint8, offset=0)
+    labels = labels.reshape(-1, img_size+1)
+    labels = labels.T
+    print("Done")
+
+    return labels[0]
+
+def _load_img(file_name):
+    file_path = dataset_dir + "/" + file_name
+
+    print("Converting " + file_name + " to NumPy Array ...")
+    with gzip.open(file_path, 'rb') as f:
+            data = np.frombuffer(f.read(), np.uint8, offset=0)
+    data = data.reshape(-1, img_size+1)
+    data = np.delete(data, 0, 1)
+    print("Done")
+
+    return data
+
+def _convert_numpy():
+    dataset = {}
+    dataset['train_img'] =  _load_img(key_file['train'])
+    dataset['train_label'] = _load_label(key_file['train'])
+    dataset['test_img'] = _load_img(key_file['test'])
+    dataset['test_label'] = _load_label(key_file['test'])
+
+    return dataset
+
+def init_cifar10():
+    dataset = _convert_numpy()
+    print("Creating pickle file ...")
+    with open(save_file, 'wb') as f:
+        pickle.dump(dataset, f, -1)
+    print("Done!")
+
+def _change_one_hot_label(X):
+    T = np.zeros((X.size, 10))
+    for idx, row in enumerate(T):
+        row[X[idx]] = 1
+
+    return T
+
+def load_cifar10(normalize=True, flatten=True, one_hot_label=False):
+    """CIFAR-10データセットの読み込み
+
+    Parameters
+    ----------
+    normalize : 画像のピクセル値を0.0~1.0に正規化する
+    one_hot_label :
+        one_hot_labelがTrueの場合、ラベルはone-hot配列として返す
+        one-hot配列とは、たとえば[0,0,1,0,0,0,0,0,0,0]のような配列
+    flatten : 画像を一次元配列に平にするかどうか
+
+    Returns
+    -------
+    (訓練画像, 訓練ラベル), (テスト画像, テストラベル)
+    """
+    if not os.path.exists(save_file):
+        init_cifar10()
+
+    with open(save_file, 'rb') as f:
+        dataset = pickle.load(f)
+
+    if normalize:
+        for key in ('train_img', 'test_img'):
+            dataset[key] = dataset[key].astype(np.float32)
+            dataset[key] /= 255.0
+
+    if one_hot_label:
+        dataset['train_label'] = _change_one_hot_label(dataset['train_label'])
+        dataset['test_label'] = _change_one_hot_label(dataset['test_label'])
+
+    if not flatten:
+         for key in ('train_img', 'test_img'):
+            dataset[key] = dataset[key].reshape(-1, 3, 32, 32)
+
+    return (dataset['train_img'], dataset['train_label']), (dataset['test_img'], dataset['test_label'])
+
+
+if __name__ == '__main__':
+    init_cifar10()

+# coding: utf-8
+#import cupy as cp
+import numpy as cp
+import numpy as np
+
+def cross_entropy_error(y, t):
+    if y.ndim == 1:
+        t = t.reshape(1, t.size)
+        y = y.reshape(1, y.size)
+
+    # 教師データがone-hot-vectorの場合、正解ラベルのインデックスに変換
+    if t.size == y.size:
+        t = t.argmax(axis=1)
+
+    batch_size = y.shape[0]
+    return -cp.sum(cp.log(y[cp.arange(batch_size), t])) / batch_size

+# coding: utf-8
+#import cupy as cp
+import numpy as cp
+import numpy as np
+from functions import *
+from util import im2col, col2im, DW_im2col
+
+
+class Relu:
+    def __init__(self):
+        self.mask = None
+
+    def forward(self, x):
+        self.mask = (x <= 0)
+        out = x.copy()
+        out[self.mask] = 0
+
+        return out
+
+    def backward(self, dout):
+        dout[self.mask] = 0
+        dx = dout
+
+        return dx
+
+
+class SoftmaxWithLoss:
+    def __init__(self):
+        self.loss = None
+        self.y = None # softmaxの出力
+        self.t = None # 教師データ
+
+    def forward(self, x, t):
+        self.t = t
+        self.y = softmax(x)
+        self.loss = cross_entropy_error(self.y, self.t)
+
+        return self.loss
+
+    def backward(self, dout=1):
+        batch_size = self.t.shape[0]
+        if self.t.size == self.y.size: # 教師データがone-hot-vectorの場合
+            dx = (self.y - self.t) / batch_size
+        else:
+            dx = self.y.copy()
+            dx[np.arange(batch_size), self.t] -= 1
+            dx = dx / batch_size
+
+        return dx
+
+class LightNormalization:
+    """
+    """
+    def __init__(self, momentum=0.9, running_mean=None, running_var=None):
+        self.momentum = momentum
+        self.input_shape = None # Conv層の場合は4次元、全結合層の場合は2次元
+
+        # テスト時に使用する平均と分散
+        self.running_mean = running_mean
+        self.running_var = running_var
+
+        # backward時に使用する中間データ
+        self.batch_size = None
+        self.xc = None
+        self.std = None
+
+    def forward(self, x, train_flg=True):
+        self.input_shape = x.shape
+        if x.ndim == 2:
+            N, D = x.shape
+            x = x.reshape(N, D, 1, 1)
+
+        x = x.transpose(0, 2, 3, 1)
+        out = self.__forward(x, train_flg)
+        out = out.transpose(0, 3, 1, 2)
+
+        return out.reshape(*self.input_shape)
+
+    def __forward(self, x, train_flg):
+        if self.running_mean is None:
+            N, H, W, C = x.shape
+            self.running_mean = cp.zeros(C, dtype=np.float32)
+            self.running_var = cp.zeros(C, dtype=np.float32)
+
+        if train_flg:
+            mu = x.mean(axis=(0, 1, 2))
+            xc = x - mu
+            var = cp.mean(xc**2, axis=(0, 1, 2), dtype=np.float32)
+            std = cp.sqrt(var + 10e-7, dtype=np.float32)
+            xn = xc / std
+
+            self.batch_size = x.shape[0]
+            self.xc = xc
+            self.xn = xn
+            self.std = std
+            self.running_mean = self.momentum * self.running_mean + (1-self.momentum) * mu
+            self.running_var = self.momentum * self.running_var + (1-self.momentum) * var
+        else:
+            xc = x - self.running_mean
+            xn = xc / ((cp.sqrt(self.running_var + 10e-7, dtype=np.float32)))
+
+        out = xn
+        return out
+
+    def backward(self, dout):
+        if dout.ndim == 2:
+            N, D = dout.shape
+            dout = dout.reshape(N, D, 1, 1)
+
+        dout = dout.transpose(0, 2, 3, 1)
+        dx = self.__backward(dout)
+        dx = dx.transpose(0, 3, 1, 2)
+
+        dx = dx.reshape(*self.input_shape)
+        return dx
+
+    def __backward(self, dout):
+        dxn = dout
+        dxc = dxn / self.std
+        dstd = -cp.sum((dxn * self.xc) / (self.std * self.std), axis=0)
+        dvar = 0.5 * dstd / self.std
+        dxc += (2.0 / self.batch_size) * self.xc * dvar
+        dmu = cp.sum(dxc, axis=0)
+        dx = dxc - dmu / self.batch_size
+
+        return dx
+
+
+
+class Convolution:
+    def __init__(self, W, stride=1, pad=0):
+        self.W = W
+        self.stride = stride
+        self.pad = pad
+
+        self.x = None
+        self.col = None
+        self.col_W = None
+
+        self.dW = None
+
+    def forward(self, x):
+        FN, C, FH, FW = self.W.shape
+        N, C, H, W = x.shape
+        out_h = 1 + int((H + 2*self.pad - FH) / self.stride)
+        out_w = 1 + int((W + 2*self.pad - FW) / self.stride)
+
+        col = im2col(x, FH, FW, self.stride, self.pad)
+        col_W = self.W.reshape(FN, -1).T
+
+        out = cp.dot(col, col_W)
+        out = out.reshape(N, out_h, out_w, -1).transpose(0, 3, 1, 2)
+
+        self.x = x
+        self.col = col
+        self.col_W = col_W
+
+        return out
+
+    def backward(self, dout):
+        FN, C, FH, FW = self.W.shape
+        dout = dout.transpose(0,2,3,1).reshape(-1, FN)
+
+        self.dW = cp.dot(self.col.T, dout)
+        self.dW = self.dW.transpose(1, 0).reshape(FN, C, FH, FW)
+
+        dcol = cp.dot(dout, self.col_W.T)
+        dx = col2im(dcol, self.x.shape, FH, FW, self.stride, self.pad)
+
+        return dx
+
+
+class Pooling:
+    def __init__(self, pool_h, pool_w, stride=1, pad=0):
+        self.pool_h = pool_h
+        self.pool_w = pool_w
+        self.stride = stride
+        self.pad = pad
+
+        self.x = None
+        self.arg_max = None
+
+    def forward(self, x):
+        N, C, H, W = x.shape
+        out_h = int(1 + (H - self.pool_h) / self.stride)
+        out_w = int(1 + (W - self.pool_w) / self.stride)
+
+        col = im2col(x, self.pool_h, self.pool_w, self.stride, self.pad)
+        col = col.reshape(-1, self.pool_h*self.pool_w)
+
+        arg_max = cp.argmax(col, axis=1)
+        out = cp.array(cp.max(col, axis=1), dtype=np.float32)
+        out = out.reshape(N, out_h, out_w, C).transpose(0, 3, 1, 2)
+
+        self.x = x
+        self.arg_max = arg_max
+
+        return out
+
+    def backward(self, dout):
+        dout = dout.transpose(0, 2, 3, 1)
+
+        pool_size = self.pool_h * self.pool_w
+        dmax = cp.zeros((dout.size, pool_size), dtype=np.float32)
+        dmax[cp.arange(self.arg_max.size), self.arg_max.flatten()] = dout.flatten()
+        dmax = dmax.reshape(dout.shape + (pool_size,))
+
+        dcol = dmax.reshape(dmax.shape[0] * dmax.shape[1] * dmax.shape[2], -1)
+        dx = col2im(dcol, self.x.shape, self.pool_h, self.pool_w, self.stride, self.pad)
+
+        return dx
+
+class DW_Convolution:
+    def __init__(self, W, stride=1, pad=0):
+        self.W = W
+        self.stride = stride
+        self.pad = pad
+
+        self.x = None
+        self.col = None
+        self.col_W = None
+
+        self.dW = None
+        self.db = None
+
+
+
+    def forward(self, x):
+        FN, C, FH, FW = self.W.shape
+        N, C, H, W = x.shape
+        out_h = 1 + int((H + 2*self.pad - FH) / self.stride)
+        out_w = 1 + int((W + 2*self.pad - FW) / self.stride)
+
+        col = DW_im2col(x, FH, FW, self.stride, self.pad)
+        col_W = self.W.reshape(FN, -1).T
+
+        outlist = []
+        outlist = np.zeros((FN, N*H*W, 1))
+        for count in range(FN):
+            outlist[count] = np.dot(col[count, :, :], col_W[:, count]).reshape(-1,1)
+
+        out = outlist.transpose(1,0,2)
+        out = out.reshape(N, out_h, out_w, -1).transpose(0, 3, 1, 2)
+
+        self.x = x
+        self.col = col
+        self.col_W = col_W
+        return out
+
+
+    def backward(self, dout):
+        FN, C, FH, FW = self.W.shape
+        N, XC, H, W = dout.shape
+        dout = dout.transpose(0,2,3,1).reshape(-1, FN)
+
+
+        dW_list = np.zeros((FN, FH*FW))
+        dcol_list = np.zeros((N * H * W, FN, FH * FW))
+        for count in range(FN):
+            dW_list[count] = np.dot(self.col[count].transpose(1,0), dout[:, count])
+            dcol_list[:,count,:] = np.dot(dout[:,count].reshape(-1,1), self.col_W.T[count,:].reshape(1,-1))
+        self.dW = dW_list
+        self.dW = self.dW.reshape(FN, C, FH, FW)
+
+
+        dcol = dcol_list
+        dx = col2im(dcol, self.x.shape, FH, FW, self.stride, self.pad)
+
+
+        return dx
+
+class Affine:
+    def __init__(self, W):
+        self.W =W
+#        self.b = b
+
+        self.x = None
+        self.original_x_shape = None
+        #self.dW = None
+#        self.db = None
+
+    def forward(self, x):
+        self.original_x_shape = x.shape
+        x = x.reshape(x.shape[0], -1)
+        self.x = x
+
+        out = cp.dot(self.x, self.W) #+ self.b
+
+        return out
+
+    def backward(self, dout):
+        dx = cp.dot(dout, self.W.T)
+        self.dW = cp.dot(self.x.T, dout)
+#        self.db = cp.sum(dout, axis=0)
+
+        dx = dx.reshape(*self.original_x_shape)  #  return dx

+import sys, os
+sys.path.append(os.pardir)
+import pickle
+import numpy as cp
+import numpy as np
+from collections import OrderedDict
+from layers import *
+
+
+class SimpleConvNet:
+    def __init__(self, input_dim=(3, 32, 32),
+                 conv_param={'filter_num':(32, 32, 64), 'filter_size':3, 'pad':1, 'stride':1},
+                 hidden_size=512, output_size=10, weight_init_std=0.01, pretrained=False):
+        filter_num = conv_param['filter_num']
+        filter_size = conv_param['filter_size']
+        filter_pad = conv_param['pad']
+        filter_stride = conv_param['stride']
+        input_size = input_dim[1]
+        conv_output_size = (input_size - filter_size + 2*filter_pad) / filter_stride + 1
+        conv_data_size    = int(filter_num[0] *  conv_output_size    *  conv_output_size   )
+        pool1_output_size = int(filter_num[1] * (conv_output_size/2) * (conv_output_size/2))
+        pool2_output_size = int(filter_num[2] * (conv_output_size/4) * (conv_output_size/4))
+        pool3_output_size = int(filter_num[2] * (conv_output_size/8) * (conv_output_size/8))
+
+        self.params = {}
+        if pretrained:
+            weights = self.load_weights()
+            self.params['W1'] = cp.array(weights['W1'], dtype=np.float32)
+            self.params['W2'] = cp.array(weights['W2'], dtype=np.float32)
+            self.params['W3'] = cp.array(weights['W3'], dtype=np.float32)
+            self.params['W4'] = cp.array(weights['W4'], dtype=np.float32)
+            self.params['W5'] = cp.array(weights['W5'], dtype=np.float32)
+            self.params['W6'] = cp.array(weights['W6'], dtype=np.float32)
+            self.params['W7'] = cp.array(weights['W7'], dtype=np.float32)
+        else:
+            self.params['W1'] = cp.array( weight_init_std * \
+                                cp.random.randn(filter_num[0], input_dim[0], filter_size, filter_size), dtype=np.float32)
+
+            self.params['W2'] = cp.array( weight_init_std * \
+                                cp.random.randn(filter_num[1], filter_num[0], 1, 1), dtype=np.float32)
+
+            self.params['W3'] = cp.array( weight_init_std * \
+                                cp.random.randn(filter_num[1], 1, filter_size, filter_size), dtype=np.float32)
+
+            self.params['W4'] = cp.array( weight_init_std * \
+                                cp.random.randn(filter_num[2], filter_num[1], 1, 1), dtype=np.float32)
+
+            self.params['W5'] = cp.array( weight_init_std * \
+                                cp.random.randn(filter_num[2], 1, filter_size, filter_size), dtype=np.float32)
+
+            self.params['W6'] = cp.array( weight_init_std * \
+                                cp.random.randn(pool3_output_size, hidden_size), dtype=np.float32)
+
+            self.params['W7'] = cp.array( weight_init_std * \
+                                cp.random.randn(hidden_size, output_size), dtype=np.float32)
+
+        self.layers = OrderedDict()
+        self.layers['Conv1'] = Convolution(self.params['W1'],
+                                           conv_param['stride'], conv_param['pad'])
+        self.layers['LightNorm1'] = LightNormalization()
+        self.layers['Relu1'] = Relu()
+        self.layers['Pool1'] = Pooling(pool_h=2, pool_w=2, stride=2)
+
+        self.layers['Conv2'] = Convolution(self.params['W2'],
+                                           1, 0)
+        self.layers['LightNorm2'] = LightNormalization()
+        self.layers['Relu2'] = Relu()
+        self.layers['Conv3'] = DW_Convolution(self.params['W3'],
+                                           conv_param['stride'], conv_param['pad'])
+        self.layers['LightNorm3'] = LightNormalization()
+        self.layers['Relu3'] = Relu()
+        self.layers['Pool2'] = Pooling(pool_h=2, pool_w=2, stride=2)
+
+        self.layers['Conv4'] = Convolution(self.params['W4'],
+                                           1, 0)
+        self.layers['LightNorm4'] = LightNormalization()
+        self.layers['Relu4'] = Relu()
+        self.layers['Conv5'] = DW_Convolution(self.params['W5'],
+                                           conv_param['stride'], conv_param['pad'])
+        self.layers['LightNorm5'] = LightNormalization()
+        self.layers['Relu5'] = Relu()
+        self.layers['Pool3'] = Pooling(pool_h=2, pool_w=2, stride=2)
+
+        self.layers['Affine4'] = Affine(self.params['W6'])
+        self.layers['LightNorm6'] = LightNormalization()
+        self.layers['Relu6'] = Relu()
+
+        self.layers['Affine5'] = Affine(self.params['W7'])
+
+        self.last_layer = SoftmaxWithLoss()
+
+    def predict(self, x):
+        for layer in self.layers.values():
+            x = layer.forward(x)
+        return x
+
+    def accuracy(self, x, t, batch_size=100):
+        if t.ndim != 1 : t = np.argmax(t, axis=1)
+
+        acc = 0.0
+
+        for i in range(int(x.shape[0] / batch_size)):
+            tx = x[i*batch_size:(i+1)*batch_size]
+            tt = t[i*batch_size:(i+1)*batch_size]
+            y = self.predict(tx)
+            y = np.argmax(y, axis=1)
+            print("answer : ", tt)
+            print("predict : ", y)
+            acc += np.sum(y == tt) #numpy
+
+        return acc / x.shape[0]
+
+    def gradient(self, x, t):
+        self.loss(x, t)
+
+        dout = 1
+        dout = self.last_layer.backward(dout)
+
+        layers = list(self.layers.values())
+        layers.reverse()
+        for layer in layers:
+            dout = layer.backward(dout)
+
+        grads = {}
+        grads['W1'] = self.layers['Conv1'].dW
+        grads['W2'] = self.layers['Conv2'].dW
+        grads['W3'] = self.layers['Conv3'].dW
+        grads['W4'] = self.layers['Conv4'].dW
+        grads['W5'] = self.layers['Conv5'].dW
+        grads['W6'] = self.layers['Affine4'].dW
+        grads['W7'] = self.layers['Affine5'].dW
+        return grads
+
+    # ���� ����ġ �ҷ����� / SimpleconvNet�� pretrained ���� �߰��� : True�� ����ġ �о� ����
+    def load_weights(self, file_name='params.pkl'):
+        weights = []
+        with open(file_name, 'rb') as f:
+            weights = pickle.load(f)
+            return weights

+from simple_convnet4 import *
+from dataset.cifar10 import load_cifar10
+
+def batch_(data, lbl, pre, size = 100):
+   return data[pre: pre+size], lbl[pre: pre+size]
+
+network = SimpleConvNet(input_dim=(3,32,32),
+                        conv_param = {'filter_num': (32, 32, 64), 'filter_size': 3, 'pad': 1, 'stride': 1},
+                        hidden_size=512, output_size=10, weight_init_std=0.01, pretrained=True)
+
+(x_train, t_train), (x_test, t_test) = load_cifar10(flatten=False)
+
+print("Length of test data: ",len(x_test))
+
+batch_size = 100
+epoch = int(len(x_test) / batch_size)
+acc = 0
+for i in range(epoch):
+   t_img, t_lbl = batch_(x_test, t_test, i*batch_size, batch_size)
+   t = network.accuracy(t_img, t_lbl, batch_size)
+   acc += t * batch_size
+
+print("Accuracy : ",str(acc / len(x_test)*100),'%')

+# coding: utf-8
+#import cupy as cp
+import numpy as cp
+import numpy as np
+
+def DW_im2col(input_data, filter_h, filter_w, stride=1, pad=0):
+    """다수의 이미지를 입력받아 2차원 배열로 변환한다(평탄화).
+
+    Parameters
+    ----------
+    input_data : 4차원 배열 형태의 입력 데이터(이미지 수, 채널 수, 높이, 너비)
+    filter_h : 필터의 높이
+    filter_w : 필터의 너비
+    stride : 스트라이드
+    pad : 패딩
+
+    Returns
+    -------
+    col : 2차원 배열
+    """
+    N, C, H, W = input_data.shape
+    out_h = (H + 2 * pad - filter_h) // stride + 1
+    out_w = (W + 2 * pad - filter_w) // stride + 1
+
+    img = np.pad(input_data, [(0, 0), (0, 0), (pad, pad), (pad, pad)], 'constant')
+    col = np.zeros((N, C, filter_h, filter_w, out_h, out_w))
+
+    for y in range(filter_h):
+        y_max = y + stride * out_h
+        for x in range(filter_w):
+            x_max = x + stride * out_w
+            col[:, :, y, x, :, :] = img[:, :, y:y_max:stride, x:x_max:stride]
+
+    col = col.transpose(1, 0, 4, 5, 2, 3).reshape(C, N * out_h * out_w, -1)
+    return col
+
+
+def im2col(input_data, filter_h, filter_w, stride=1, pad=0):
+    """
+
+    Parameters
+    ----------
+    input_data : (データ数, チャンネル, 高さ, 幅)の4次元配列からなる入力データ
+    filter_h : フィルターの高さ
+    filter_w : フィルターの幅
+    stride : ストライド
+    pad : パディング
+
+    Returns
+    -------
+    col : 2次元配列
+    """
+    N, C, H, W = input_data.shape
+    out_h = (H + 2*pad - filter_h)//stride + 1
+    out_w = (W + 2*pad - filter_w)//stride + 1
+
+    img = cp.pad(input_data, [(0,0), (0,0), (pad, pad), (pad, pad)], 'constant')
+    col = cp.zeros((N, C, filter_h, filter_w, out_h, out_w), dtype=np.float32)
+
+    for y in range(filter_h):
+        y_max = y + stride*out_h
+        for x in range(filter_w):
+            x_max = x + stride*out_w
+            col[:, :, y, x, :, :] = img[:, :, y:y_max:stride, x:x_max:stride]
+
+    col = col.transpose(0, 4, 5, 1, 2, 3).reshape(N*out_h*out_w, -1)
+    return col
+
+
+def col2im(col, input_shape, filter_h, filter_w, stride=1, pad=0):
+    """
+
+    Parameters
+    ----------
+    col :
+    input_shape : 入力データの形状（例：(10, 1, 28, 28)）
+    filter_h :
+    filter_w
+    stride
+    pad
+
+    Returns
+    -------
+
+    """
+    N, C, H, W = input_shape
+    out_h = (H + 2*pad - filter_h)//stride + 1
+    out_w = (W + 2*pad - filter_w)//stride + 1
+    col = col.reshape(N, out_h, out_w, C, filter_h, filter_w).transpose(0, 3, 4, 5, 1, 2)
+
+    img = cp.zeros((N, C, H + 2*pad + stride - 1, W + 2*pad + stride - 1), dtype=np.float32)
+    for y in range(filter_h):
+        y_max = y + stride*out_h
+        for x in range(filter_w):
+            x_max = x + stride*out_w
+            img[:, :, y:y_max:stride, x:x_max:stride] += col[:, :, y, x, :, :]
+
+    return img[:, :, pad:H + pad, pad:W + pad]