如何解决多输出模型的一个图像输入：InvalidArgumentError
我有一个项目可以从图像中预测 8 个输出（4 个坐标），当我运行 model.fit() 时发生此错误。
我的 .csv 文件有 768 个数据，例如：
name x1 y1 x2 y2 x3 y3 x4 y4
0 frame220.png 254 720 619 432 1189 720 641 432
1 frame570.png 230 720 641 432 1154 720 686 432
....
df = pd.read_csv('./dataFile/data.csv')
我的模型（VGG16）：
def build_model():
input = Input(shape = (224,224,3),batch_size=32)
x = Conv2D(64,(3,padding= 'same',activation= 'relu')(input)
x = Conv2D(64,activation= 'relu')(x)
x = MaxPooling2D(pool_size=(2,2),strides=(2,2))(x)
x = Conv2D(128,activation= 'relu')(x)
x = Conv2D(128,2))(x)
x = Conv2D(256,activation= 'relu')(x)
x = Conv2D(256,2))(x)
x = Conv2D(512,activation= 'relu')(x)
x = Conv2D(512,2))(x)
x = Flatten()(x)
x = Dense(256,activation= 'relu')(x)
x = Dropout(0.5)(x)
x1_dence = Dense(128,activation='softmax')(x)
x1 = Dense(1,name = 'x1')(x)
y1_dence = Dense(128,activation='softmax')(x)
y1 = Dense(1,name = 'y1')(x)
x2_dence = Dense(128,activation='softmax')(x)
x2 = Dense(1,name = 'x2')(x)
y2_dence = Dense(128,activation='softmax')(x)
y2 = Dense(1,name = 'y2')(x)
x3_dence = Dense(128,activation='softmax')(x)
x3 = Dense(1,name = 'x3')(x)
y3_dence = Dense(128,activation='softmax')(x)
y3 = Dense(1,name = 'y3')(x)
x4_dence = Dense(128,activation='softmax')(x)
x4 = Dense(1,name = 'x4')(x)
y4_dence = Dense(128,activation='softmax')(x)
y4 = Dense(1,name = 'y4')(x)
model = Model( inputs = input,outputs = [x1,y1,x2,y2,x3,y3,x4,y4])
return model
模型总结
Model: "model_6"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_7 (InputLayer) [(32,3)] 0
__________________________________________________________________________________________________
conv2d_78 (Conv2D) (32,64) 1792 input_7[0][0]
__________________________________________________________________________________________________
conv2d_79 (Conv2D) (32,64) 36928 conv2d_78[0][0]
__________________________________________________________________________________________________
max_pooling2d_30 (MaxPooling2D) (32,112,64) 0 conv2d_79[0][0]
__________________________________________________________________________________________________
conv2d_80 (Conv2D) (32,128) 73856 max_pooling2d_30[0][0]
__________________________________________________________________________________________________
conv2d_81 (Conv2D) (32,128) 147584 conv2d_80[0][0]
__________________________________________________________________________________________________
max_pooling2d_31 (MaxPooling2D) (32,56,128) 0 conv2d_81[0][0]
__________________________________________________________________________________________________
conv2d_82 (Conv2D) (32,256) 295168 max_pooling2d_31[0][0]
__________________________________________________________________________________________________
conv2d_83 (Conv2D) (32,256) 590080 conv2d_82[0][0]
__________________________________________________________________________________________________
conv2d_84 (Conv2D) (32,256) 590080 conv2d_83[0][0]
__________________________________________________________________________________________________
max_pooling2d_32 (MaxPooling2D) (32,28,256) 0 conv2d_84[0][0]
__________________________________________________________________________________________________
conv2d_85 (Conv2D) (32,512) 1180160 max_pooling2d_32[0][0]
__________________________________________________________________________________________________
conv2d_86 (Conv2D) (32,512) 2359808 conv2d_85[0][0]
__________________________________________________________________________________________________
conv2d_87 (Conv2D) (32,512) 2359808 conv2d_86[0][0]
__________________________________________________________________________________________________
max_pooling2d_33 (MaxPooling2D) (32,14,512) 0 conv2d_87[0][0]
__________________________________________________________________________________________________
conv2d_88 (Conv2D) (32,512) 2359808 max_pooling2d_33[0][0]
__________________________________________________________________________________________________
conv2d_89 (Conv2D) (32,512) 2359808 conv2d_88[0][0]
__________________________________________________________________________________________________
conv2d_90 (Conv2D) (32,512) 2359808 conv2d_89[0][0]
__________________________________________________________________________________________________
max_pooling2d_34 (MaxPooling2D) (32,7,512) 0 conv2d_90[0][0]
__________________________________________________________________________________________________
flatten_6 (Flatten) (32,25088) 0 max_pooling2d_34[0][0]
__________________________________________________________________________________________________
dense_54 (Dense) (32,256) 6422784 flatten_6[0][0]
__________________________________________________________________________________________________
dropout_6 (Dropout) (32,256) 0 dense_54[0][0]
__________________________________________________________________________________________________
x1 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
y1 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
x2 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
y2 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
x3 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
y3 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
x4 (Dense) (32,1) 257 dropout_6[0][0]
__________________________________________________________________________________________________
y4 (Dense) (32,1) 257 dropout_6[0][0]
==================================================================================================
Total params: 21,139,528
Trainable params: 21,528
Non-trainable params: 0
将图像文件夹连接到 CSV 文件
train_image = []
for i in tqdm(range(df.shape[0])):
img = image.load_img('./dataframe/frame' + str(i)+ '.png',target_size=(224,3))
img = image.img_to_array(img)
train_image.append(img)
X = np.array(train_image)
print(X.shape)
y = np.array(df.drop(['name'],axis=1))
print(y.shape)
100%|██████████| 786/786 [05:07<00:00,2.55it/s](786,3)
(786,8)
拆分数据 - 我在 X_train 中展示了一个样本 > 结果是图像
X_train,X_test,y_train,y_test = train_test_split(X,y,random_state=42,test_size=0.2)
y_train[67]
array([ 279,720,643,432,1217,672,432])
model = build_model()
optimizer = tf.keras.optimizers.SGD(learning_rate=0.001)
model.compile(
optimizer = optimizer,loss={'x1': 'mse','y1': 'mse','x2': 'mse','y2': 'mse','x3': 'mse','y3': 'mse','x4': 'mse','y4': 'mse'},metrics={
'x1': tf.keras.metrics.RootMeanSquaredError(),'y1': tf.keras.metrics.RootMeanSquaredError(),'x2': tf.keras.metrics.RootMeanSquaredError(),'y2': tf.keras.metrics.RootMeanSquaredError(),'x3': tf.keras.metrics.RootMeanSquaredError(),'y3': tf.keras.metrics.RootMeanSquaredError(),'x4': tf.keras.metrics.RootMeanSquaredError(),'y4': tf.keras.metrics.RootMeanSquaredError()
}
)
最后当我运行 model.fit()
history = model.fit(X_train,epochs=10,batch_size=32,validation_data=(X_test,y_test))
Epoch 1/10
19/20 [===========================>..] - ETA: 0s - loss: nan - x1_loss: nan - y1_loss: nan - x2_loss: nan - y2_loss: nan - x3_loss: nan - y3_loss: nan - x4_loss: nan - y4_loss: nan - x1_root_mean_squared_error: nan - y1_root_mean_squared_error: nan - x2_root_mean_squared_error: nan - y2_root_mean_squared_error: nan - x3_root_mean_squared_error: nan - y3_root_mean_squared_error: nan - x4_root_mean_squared_error: nan - y4_root_mean_squared_error: nan
---------------------------------------------------------------------------
InvalidArgumentError Traceback (most recent call last)
<ipython-input-23-5984067494bc> in <module>()
----> 1 history = model.fit(X_train,y_test))
6 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,num_outputs)
61 except core._NotOkStatusException as e:
62 if name is not None:
InvalidArgumentError: required broadcastable shapes at loc(unknown)
[[node model_4/dropout_4/dropout/Mul_1 (defined at /usr/local/lib/python3.7/dist-packages/keras/layers/core.py:205) ]] [Op:__inference_train_function_10022]
Function call stack:
train_function
那么，我错在哪里？以及如何修复它？谢谢你
版权声明：本文内容由互联网用户自发贡献，该文观点与技术仅代表作者本人。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容， 请发送邮件至 dio@foxmail.com 举报，一经查实，本站将立刻删除。
