ValueError ResNet Keras

Question

I was trying to use the Keras ResNet 50 application model on my problem with this code:

#Tensorflow and tf.keras
import tensorflow as tf
from tensorflow import keras
#tf.enable_eager_execution()

#Helper libraries
import numpy as np
import matplotlib.pyplot as plt

import Muenzbetragserkennung_input_ResNet

#print(tf.__version__)

#Dataset
#Training and test data
(train_images, train_labels), (test_images, test_labels) = 
Muenzbetragserkennung_input_ResNet.read_input_shuffle()

batch_size, height, width, channels = train_images.shape
train_images = train_images / 255.0
test_images = test_images / 255.0
print(train_images.shape)

#Build the model
model = tf.keras.applications.resnet50.ResNet50(include_top=False, 
weights=None, input_tensor=None, input_shape=(height, width, channels), 
pooling='max')

model.compile(optimizer=tf.train.AdamOptimizer(),
          loss='mean_squared_error',
          metrics=['accuracy'])

#model.summary()

#Train
model.fit(train_images, train_labels, epochs=10)
#model.save_weights('models/muenzen.h5')

#Evaluate
loss, accuracy = model.evaluate(test_images, test_labels)
print('Accuracy', accuracy)

#Prediction
prediction = model.predict(test_images[0:1])
print(prediction)

But got the following Ouput/Error:

Shape train images: (3865, 240, 320, 3)

Shape train labels: (3865,)

Shape test images: (967, 240, 320, 3)

Shape test labels: (967,)

(3865, 240, 320, 3)

Traceback (most recent call last):
File"C:/Users/Christian/PycharmProjects/MuenzbetragserkennungResNet/Muenzbetragserkennung_ResNet.py", line 34, in model.fit(train_images, train_labels, epochs=10)

File "C:\Users\Christian\AppData\Roaming\Python\Python36\site-packages\tensorflow\python\keras\engine\training.py", line 1278, in fit validation_split=validation_split)

File "C:\Users\Christian\AppData\Roaming\Python\Python36\site-packages\tensorflow\python\keras\engine\training.py", line 917, in _standardize_user_data exception_prefix='target')

File "C:\Users\Christian\AppData\Roaming\Python\Python36\site-packages\tensorflow\python\keras\engine\training_utils.py", line 191, in standardize_input_data ' but got array with shape ' + str(data_shape))

ValueError: Error when checking target: expected global_max_pooling2d to have shape (2048,) but got array with shape (1,)

Process finished with exit code 1

I already tried different pooling versions, but only got other ValueErrors. The model should output one value (worth of coins in the image).

Thank you in advance for your help.

Irrelevant to your issue, but you are mixing a regression loss (MSE) with a classification metric (accuracy), which should not be the case. — desertnaut
I know, the problem is: The input is an image with, for example, 20 coins. I want to know how much the 20 coins are worth in total. If I get the output of 16.10€ it is closer to the real value (e.g., 16.20€) than 7.50€. Nevertheless, I want to know for how many images the value is correct (e.g., for this image 16.20€). — chris

Dr. Snoopy Dr. Snoopy · Accepted Answer · 2018-09-02T14:18:02

The problem is that your labels are one-dimensional, but the output of your model is a 2048-dimensional vector. This is natural as you didn't add any layers to produce the proper output. This can be done as:

resnet_model = tf.keras.applications.resnet50.ResNet50(include_top=False, 
weights=None, input_tensor=None, input_shape=(height, width, channels), 
pooling='max')

x = Dense(128, activation='relu')(resnet_model.output)
x = Dense(1, activation='relu')(x)

model = Model(resnet_model.input, x)

Note that the last Dense layer outputs a single scalar, which is now compatible with your targets.

ValueError ResNet Keras

1 Answers