英文:
Facing input shape/layer problem in keras 2D convutional network
问题
我面临以下错误：
数值错误：调用层 'conv2d'（类型 Conv2D）时遇到异常。
由于从 1 减去 3 导致负维度大小，用于 'sequential/conv2d/Conv2D' 节点（输入形状为：[?,200,1,1], [3,3,1,9]）。
层 'conv2d'（类型 Conv2D）接收到的调用参数：
• inputs=tf.Tensor(shape=(None, 200, 1, 1), dtype=float32)
我的模型：
model = Sequential()
model.add(Conv2D(9, kernel_size=3, activation='relu', input_shape=(200,200, 1)))
model.add(Conv2D(4, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(2, activation='sigmoid'))
model.predict(image)
这里的 "image" 是一个维度为 (200, 200, 1) 的数组，其中包含了 0 和 1 的值。
我该怎么办？
英文:
I face the following error:
ValueError: Exception encountered when calling layer 'conv2d' (type Conv2D).
Negative dimension size caused by subtracting 3 from 1 for '{{node sequential/conv2d/Conv2D}} = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], explicit_paddings=[], padding="VALID", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true](sequential/Cast, sequential/conv2d/Conv2D/ReadVariableOp)' with input shapes: [?,200,1,1], [3,3,1,9].
Call arguments received by layer 'conv2d' (type Conv2D):
• inputs=tf.Tensor(shape=(None, 200, 1, 1), dtype=float32)
my model:
model = Sequential()
model.add(Conv2D(9, kernel_size=3, activation='relu', input_shape=(200,200, 1)))
model.add(Conv2D(4, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(2, activation='sigmoid'))
model.predict(image)
Here "image" is a array of dimensions (200, 200, 1) with values in it as 0's&1's
what should I do??
答案1
得分: 1
你的模型的输入形状是 (200,200, 1)，但不要忘记训练和推理是在批次上进行的。你的真正输入形状是 (None, 200,200, 1)。
如果你只想针对单个图像进行预测，你应该在开头添加另一个维度，像这样：
image = np.expand_dims(image, 0)
# 图像形状应该是 (1, 200,200, 1)
model.predict(image)
英文:
Your model's input is (200,200, 1) but don't forget that training and inference is done on batches. Your true input shape is (None, 200,200, 1)
If you want predictions only for a single image you should add another dimension at the beginning, like this:
image=np.expand_dims(image, 0)
# image shape should be (1, 200,200, 1)
model.predict(image)
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论