我不明白我的模型能够输出模型摘要，但是当我想通过编写 model.predict 来测试它是否真的有效时，它向我显示了 Incompatible shapes 错误（如下所述） .
def PVT():
# Inputs
input = layers.Input(shape=input_shape)
augment = data_augmentation(input)
# Stage 1
patches_1 = Patch(patch_size_1)(augment)
patches_1 = PatchEncoder(num_patches=(image_size // patch_size_1) ** 2, projection_dim=projection_dim)(patches_1)
for _ in range(transformer_layers):
x1 = layers.LayerNormalization(epsilon=1e-6)(patches_1)
attention_output = layers.MultiHeadAttention(
num_heads = num_heads, key_dim = projection_dim, dropout = 0.1
)(x1,x1)
x2 = attention_output + patches_1
x3 = layers.LayerNormalization(epsilon=1e-6)(x2)
x3 = mlp(x3, transformer_units, 0.2)
encoded_patches = x3 + x2
representation = layers.LayerNormalization(epsilon=1e-6)(encoded_patches)
input_2 = layers.Reshape([image_size // patch_size_1, image_size // patch_size_1, 64])(representation)
# Stage 2
patches_2 = Patch(patch_size_2)(input_2)
patches_2 = PatchEncoder(num_patches=(image_size // patch_size_2) ** 2, projection_dim=projection_dim)(patches_2)
for _ in range(transformer_layers):
x1 = layers.LayerNormalization(epsilon=1e-6)(patches_2)
attention_output = layers.MultiHeadAttention(
num_heads = num_heads, key_dim = projection_dim, dropout = 0.1
)(x1,x1)
x2 = attention_output + patches_2
x3 = layers.LayerNormalization(epsilon=1e-6)(x2)
x3 = mlp(x3, transformer_units, 0.2)
encoded_patches = x3 + x2
encoded_patches.shape
representation = layers.LayerNormalization(epsilon=1e-6)(encoded_patches)
input_3 = layers.Reshape([image_size // patch_size_2, image_size // patch_size_2, 64])(representation)
# Stage 3
patches_3 = Patch(patch_size_3)(input_3)
patches_3 = PatchEncoder(num_patches=(image_size // patch_size_3) ** 2, projection_dim=projection_dim)(patches_3)
for _ in range(transformer_layers):
x1 = layers.LayerNormalization(epsilon=1e-6)(patches_3)
attention_output = layers.MultiHeadAttention(
num_heads = num_heads, key_dim = projection_dim, dropout = 0.1
)(x1,x1)
x2 = attention_output + patches_3
x3 = layers.LayerNormalization(epsilon=1e-6)(x2)
x3 = mlp(x3, transformer_units, 0.2)
encoded_patches = x3 + x2
representation = layers.LayerNormalization(epsilon=1e-6)(encoded_patches)
input_4 = layers.Reshape([image_size // patch_size_3, image_size // patch_size_3, 64])(representation)
# Stage 4
patches_4 = Patch(patch_size_4)(input_4)
patches_4 = PatchEncoder(num_patches=(image_size // patch_size_4) ** 2, projection_dim=projection_dim)(patches_4)
for _ in range(transformer_layers):
x1 = layers.LayerNormalization(epsilon=1e-6)(patches_4)
attention_output = layers.MultiHeadAttention(
num_heads = num_heads, key_dim = projection_dim, dropout = 0.1
)(x1,x1)
x2 = attention_output + patches_4
x3 = layers.LayerNormalization(epsilon=1e-6)(x2)
x3 = mlp(x3, transformer_units, 0.2)
encoded_patches = x3 + x2
representation = layers.LayerNormalization(epsilon=1e-6)(encoded_patches)
input_5 = layers.Reshape([image_size // patch_size_4, image_size // patch_size_4, 64])(representation)
representation = layers.Flatten()(input_5)
representation = layers.Dropout(0.5)(representation)
# Classify outputs.
logits = layers.Dense(num_classes)(representation)
Create the Keras model.
model = keras.Model(inputs=input, outputs=logits)
return model
---------------------------------------------------------------------------
InvalidArgumentError Traceback (most recent call last)
<ipython-input-20-3e4aa14aa594> in <module>()
----> 1 model.predict(xtrain)
5 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
58 ctx.ensure_initialized()
59 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
---> 60 inputs, attrs, num_outputs)
61 except core._NotOkStatusException as e:
62 if name is not None:
InvalidArgumentError: Incompatible shapes: [32,4,64] vs. [81,64]
[[node model_2/patch_encoder_11/add (defined at <ipython-input-6-93bf719690a9>:12) ]] [Op:__inference_predict_function_48830]
Function call stack:
predict_function
Model: "model_2"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_4 (InputLayer) [(None, 32, 32, 3)] 0
__________________________________________________________________________________________________
data_augmentation (Sequential) (None, 72, 72, 3) 7 input_4[0][0]
__________________________________________________________________________________________________
patch_10 (Patch) (None, None, 48) 0 data_augmentation[3][0]
__________________________________________________________________________________________________
patch_encoder_10 (PatchEncoder) (None, 324, 64) 23872 patch_10[0][0]
__________________________________________________________________________________________________
layer_normalization_124 (LayerN (None, 324, 64) 128 patch_encoder_10[0][0]
__________________________________________________________________________________________________
multi_head_attention_60 (MultiH (None, 324, 64) 82944 layer_normalization_124[0][0]
layer_normalization_124[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_119 (TFOpL (None, 324, 64) 0 multi_head_attention_60[0][0]
patch_encoder_10[0][0]
__________________________________________________________________________________________________
layer_normalization_125 (LayerN (None, 324, 64) 128 tf.__operators__.add_119[0][0]
__________________________________________________________________________________________________
dense_131 (Dense) (None, 324, 128) 8320 layer_normalization_125[0][0]
__________________________________________________________________________________________________
dropout_120 (Dropout) (None, 324, 128) 0 dense_131[0][0]
__________________________________________________________________________________________________
dense_132 (Dense) (None, 324, 64) 8256 dropout_120[0][0]
__________________________________________________________________________________________________
dropout_121 (Dropout) (None, 324, 64) 0 dense_132[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_120 (TFOpL (None, 324, 64) 0 dropout_121[0][0]
tf.__operators__.add_119[0][0]
__________________________________________________________________________________________________
layer_normalization_126 (LayerN (None, 324, 64) 128 tf.__operators__.add_120[0][0]
__________________________________________________________________________________________________
reshape_5 (Reshape) (None, 18, 18, 64) 0 layer_normalization_126[0][0]
__________________________________________________________________________________________________
patch_11 (Patch) (None, None, 4096) 0 reshape_5[0][0]
__________________________________________________________________________________________________
patch_encoder_11 (PatchEncoder) (None, 81, 64) 267392 patch_11[0][0]
__________________________________________________________________________________________________
layer_normalization_145 (LayerN (None, 81, 64) 128 patch_encoder_11[0][0]
__________________________________________________________________________________________________
multi_head_attention_70 (MultiH (None, 81, 64) 82944 layer_normalization_145[0][0]
layer_normalization_145[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_139 (TFOpL (None, 81, 64) 0 multi_head_attention_70[0][0]
patch_encoder_11[0][0]
__________________________________________________________________________________________________
layer_normalization_146 (LayerN (None, 81, 64) 128 tf.__operators__.add_139[0][0]
__________________________________________________________________________________________________
dense_152 (Dense) (None, 81, 128) 8320 layer_normalization_146[0][0]
__________________________________________________________________________________________________
dropout_140 (Dropout) (None, 81, 128) 0 dense_152[0][0]
__________________________________________________________________________________________________
dense_153 (Dense) (None, 81, 64) 8256 dropout_140[0][0]
__________________________________________________________________________________________________
dropout_141 (Dropout) (None, 81, 64) 0 dense_153[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_140 (TFOpL (None, 81, 64) 0 dropout_141[0][0]
tf.__operators__.add_139[0][0]
__________________________________________________________________________________________________
layer_normalization_147 (LayerN (None, 81, 64) 128 tf.__operators__.add_140[0][0]
__________________________________________________________________________________________________
reshape_6 (Reshape) (None, 9, 9, 64) 0 layer_normalization_147[0][0]
__________________________________________________________________________________________________
patch_12 (Patch) (None, None, 16384) 0 reshape_6[0][0]
__________________________________________________________________________________________________
patch_encoder_12 (PatchEncoder) (None, 16, 64) 1049664 patch_12[0][0]
__________________________________________________________________________________________________
layer_normalization_166 (LayerN (None, 16, 64) 128 patch_encoder_12[0][0]
__________________________________________________________________________________________________
multi_head_attention_80 (MultiH (None, 16, 64) 82944 layer_normalization_166[0][0]
layer_normalization_166[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_159 (TFOpL (None, 16, 64) 0 multi_head_attention_80[0][0]
patch_encoder_12[0][0]
__________________________________________________________________________________________________
layer_normalization_167 (LayerN (None, 16, 64) 128 tf.__operators__.add_159[0][0]
__________________________________________________________________________________________________
dense_173 (Dense) (None, 16, 128) 8320 layer_normalization_167[0][0]
__________________________________________________________________________________________________
dropout_160 (Dropout) (None, 16, 128) 0 dense_173[0][0]
__________________________________________________________________________________________________
dense_174 (Dense) (None, 16, 64) 8256 dropout_160[0][0]
__________________________________________________________________________________________________
dropout_161 (Dropout) (None, 16, 64) 0 dense_174[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_160 (TFOpL (None, 16, 64) 0 dropout_161[0][0]
tf.__operators__.add_159[0][0]
__________________________________________________________________________________________________
layer_normalization_168 (LayerN (None, 16, 64) 128 tf.__operators__.add_160[0][0]
__________________________________________________________________________________________________
reshape_7 (Reshape) (None, 4, 4, 64) 0 layer_normalization_168[0][0]
__________________________________________________________________________________________________
patch_13 (Patch) (None, None, 65536) 0 reshape_7[0][0]
__________________________________________________________________________________________________
patch_encoder_13 (PatchEncoder) (None, 4, 64) 4194624 patch_13[0][0]
__________________________________________________________________________________________________
layer_normalization_187 (LayerN (None, 4, 64) 128 patch_encoder_13[0][0]
__________________________________________________________________________________________________
multi_head_attention_90 (MultiH (None, 4, 64) 82944 layer_normalization_187[0][0]
layer_normalization_187[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_179 (TFOpL (None, 4, 64) 0 multi_head_attention_90[0][0]
patch_encoder_13[0][0]
__________________________________________________________________________________________________
layer_normalization_188 (LayerN (None, 4, 64) 128 tf.__operators__.add_179[0][0]
__________________________________________________________________________________________________
dense_194 (Dense) (None, 4, 128) 8320 layer_normalization_188[0][0]
__________________________________________________________________________________________________
dropout_180 (Dropout) (None, 4, 128) 0 dense_194[0][0]
__________________________________________________________________________________________________
dense_195 (Dense) (None, 4, 64) 8256 dropout_180[0][0]
__________________________________________________________________________________________________
dropout_181 (Dropout) (None, 4, 64) 0 dense_195[0][0]
__________________________________________________________________________________________________
tf.__operators__.add_180 (TFOpL (None, 4, 64) 0 dropout_181[0][0]
tf.__operators__.add_179[0][0]
__________________________________________________________________________________________________
layer_normalization_189 (LayerN (None, 4, 64) 128 tf.__operators__.add_180[0][0]
__________________________________________________________________________________________________
reshape_8 (Reshape) (None, 2, 2, 64) 0 layer_normalization_189[0][0]
__________________________________________________________________________________________________
flatten_2 (Flatten) (None, 256) 0 reshape_8[0][0]
__________________________________________________________________________________________________
dropout_182 (Dropout) (None, 256) 0 flatten_2[0][0]
__________________________________________________________________________________________________
dense_196 (Dense) (None, 100) 25700 dropout_182[0][0]
==================================================================================================
Total params: 5,960,875
Trainable params: 5,960,868
Non-trainable params: 7
___________________________