我想用 TensorFlow 識別圖像。 我遇到了形狀錯誤，但我該怎么辦？

Question

GmdMiss_Folder = os.path.join(os.getcwd(), '..', 'Photo', 'GMD Miss')
GmdMiss_List = os.listdir(GmdMiss_Folder)

for i in range(0, len(GmdMiss_List)):
    Img = os.path.join(os.getcwd(), GmdMiss_Folder, GmdMiss_List[i])
    Img = cv2.imread(Img, cv2.IMREAD_GRAYSCALE)
    Img = np.array(Img)
    Img = cv2.resize(Img, dsize=(1920, 1080), interpolation=cv2.INTER_AREA)
    Img_Miss_List.append(Img)
i=0

while True:
    Img = Img_Miss_List[i]
    with tf.Session() as sess:
        graph = tf.Graph()
        with graph.as_default():
            with tf.name_scope("Convolution"):
                Img = Convolution(Img)
i += 1

...
The codes below are omitted
...

def Convolution(img):
    kernel = tf.Variable(tf.truncated_normal(shape=[180, 180, 3, 3], stddev=0.1))
    img = img.astype('float32')
    img = tf.nn.conv2d(np.expand_dims(img, 0), kernel, strides=[ 1, 15, 15, 1], padding='VALID')
    return img

錯誤是..

ValueError：形狀必須為 4 級，但對於輸入形狀為 [1,1080,1920]、[180,180,3,3] 的“卷積/Conv2D”（操作：“Conv2D”）為 3 級。

Answer 1

您的conv2D kernel [filter_height, filter_width, in_channels, out_channels]期望輸入圖像有 3 個通道，並且它是一個形狀為[batch, in_height, in_width, in_channels]的 4D 張量，而您已經以灰度格式加載了輸入圖像。 因此，您需要加載具有 3 個通道的彩色圖像：

Img = cv2.imread(Img)  # By default cv2 load image in BGR format
Img = cv2.cvtColor(Img, cv2.COLOR_BGR2RGB)  # convert BGR to RGB format

希望它會有所幫助。