层“dense3”的输入 0 与层不兼容：预期 min_ndim=2，发现 ndim=1。 收到的完整形状：（4096，）

Question

I am pretty new to keras and machine learning, but my code is not working and I am pretty sure it is because of my model. I am trying to train my model with a csv file that has my images all flattened into one row. Essentially, each row of data is one image. However, this error message keeps popping up

ValueError: Exception encountered when calling layer "sequential" (type Sequential).

Input 0 of layer "dense3" is incompatible with the layer: expected min_ndim=2, found ndim=1. Full shape received: (4096,)

Call arguments received by layer "sequential" (type Sequential):
  • inputs=tf.Tensor(shape=(4096,), dtype=uint8)
  • training=False
  • mask=None

I have looked online and tried a lot of things, but I am new, so I do not completely understand everything. The image is 64x64, so I assumed that the columns were the input size.... so set the input size equal to 64 64 and also changed the predict batch size to be 64 64 that way it matches (since I am getting the real-time flattened image to be predicted). It solved the problem for the first layer (the input layer), but now the dense layers after are not working. I also have the image label at the first index of every row on the csv file.

def modelTraining():
model = keras.models.Sequential()
model.add(keras.Input(shape=(4096,)))
model.add(keras.layers.Dense(4096, activation='relu', name = "dense2"), )
model.add(keras.layers.Dense(1, activation='softmax', name = "dense3", ))

model.compile(optimizer='rmsprop',
          loss='binary_crossentropy',
          metrics=['accuracy'])

#Get our training image data
trainingImageMatrix = []
trainingLabelsArray = []
temp = 0
with open(r'C:\Users\Jason\Desktop\signLanguage\testOpenCV\training.csv', 'r', newline='') as myFile:
    reader = csv.reader(myFile, delimiter=',')
    for row in reader:
        temp+=1
        imageDataToAdd = np.zeros(4096)
        for i in range (1, 4097):
            imageDataToAdd[i - 1] = row[i]
        trainingLabelsArray.append(int(row[0])) #need to get the first index of every row
        trainingImageMatrix.append(imageDataToAdd)
        print("reading row: "+str(temp))
        #if temp >=10:
            #break
    trainingImageMatrix = np.array(trainingImageMatrix)
    trainingLabelsArray = np.array(trainingLabelsArray)
model.fit(trainingImageMatrix, trainingLabelsArray, epochs = 20, batch_size = 4096)
return model

model.predict(grayFrame.flatten(), batch_size = 4096) 

this line is in another part of my code, but this is the predict I am using atm.

Would appreciate some help and clarification on how to specify the size and how the layers should be customized for my task.

each row has the 0th index as a label and the rest is 64*64 worth of pixel brightness ---- CSV FILE IMAGE

Answer 1

I have used this dataset and converted the npy file to csv and replicated the issue. grayFrame.flatten() gives the output of 1 dimension. Adding an extra dimension will solve the error. Replace the following line

model.predict(grayFrame.flatten(), batch_size = 4096) 

with the line below

model.predict(tf.expand_dims(grayFrame.flatten(),axis=0), batch_size = 4096) 

Please find the sample code here . Thank you