TensorFlow：logits 和標簽必須具有相同的第一維，得到 logits 形狀 [120,4] 和標簽形狀 [480]

Question

I created an object detection model with hand written digits images, my model has two output layers first is to classification between digits, and second one is to detect/localize the bounding box coordinates of digit.

Output = [輸出1，輸出2]

Output1 = [0, 1, 2, 3, 4, 5, 6, 7, 8,9]（10 個單元的密集層）

Output2 = [x, y, w, h]（邊界框坐標有 4 個單位的密集層）

我面臨關於 logits 和 label 形狀的錯誤，我不明白為什么會發生此錯誤，請幫幫我。

這是csv文件標簽的SS

訓練圖像看起來像

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import tensorflow as tf
import pandas as pd

data = pd.read_csv('train_label.csv')

image = data.iloc[:,0].values
label = data.iloc[:,1:].values

train = tf.data.Dataset.from_tensor_slices((image, label))

def collector(images_file, label):
    image = tf.io.read_file('train\\'+images_file)
    image = tf.image.decode_image(image, channels=1, dtype=tf.float32)

    labels = {'label': label[0], 'coordinates': label[1:]}
    return image, labels

train = train.map(collector).batch(120)

input = tf.keras.Input(shape=(75, 75, 1))
x = tf.keras.layers.Conv2D(32, kernel_size=3, activation='relu')(input)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Conv2D(64, kernel_size=3, activation='relu')(x)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Conv2D(128, kernel_size=3, activation='relu')(x)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(128, activation='relu')(x)
output1 = tf.keras.layers.Dense(10, activation='softmax', name="label")(x)
output2 = tf.keras.layers.Dense(4, name="coordinates")(x)

model = tf.keras.Model(inputs=input, outputs=[output1, output2])

model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              optimizer='adam', metrics=['accuracy'])

model.fit(train, batch_size=120, epochs=5, verbose=1)

Epoch 1/5
Traceback (most recent call last):
  File "D:/obj_detection_and_classification/train_model.py", line 42, in <module>
    model.fit(train, batch_size=120, epochs=5, verbose=1)
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1100, in fit
    tmp_logs = self.train_function(iterator)
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\def_function.py", line 828, in __call__
    result = self._call(*args, **kwds)
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\def_function.py", line 888, in _call
    return self._stateless_fn(*args, **kwds)
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 2942, in __call__
    return graph_function._call_flat(
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 1918, in _call_flat
    return self._build_call_outputs(self._inference_function.call(
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 555, in call
    outputs = execute.execute(
  File "C:\Users\Devanshu\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\execute.py", line 59, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.InvalidArgumentError:  logits and labels must have the same first dimension, got logits shape [120,4] and labels shape [480]
     [[node sparse_categorical_crossentropy_1/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits (defined at D:/obj_detection_and_classification/train_model.py:42) ]] [Op:__inference_train_function_1524]

Function call stack:
train_function

Answer 1

僅將 model.compile 行從“SparseCategoricalCrossentropy”更改為“CategoricalCrossentropy”形式。 考慮坐標的不同損失 function，如下面的示例代碼所示。

並且，在訓練中驗證您有正確的數據輸入形狀，請參見下面示例代碼中的 model.fit(...) 行：

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import tensorflow as tf
import pandas as pd
from tensorflow import keras
from tensorflow.keras import layers

input = tf.keras.Input(shape=(75, 75, 1))
x = tf.keras.layers.Conv2D(32, kernel_size=3, activation='relu')(input)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Conv2D(64, kernel_size=3, activation='relu')(x)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Conv2D(128, kernel_size=3, activation='relu')(x)
x = tf.keras.layers.MaxPooling2D((3,3), strides=(1,1), padding='same')(x)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(128, activation='relu')(x)
output1 = tf.keras.layers.Dense(10, activation='softmax', name="label")(x)
output2 = tf.keras.layers.Dense(4, name="coordinates")(x)

model = tf.keras.Model(inputs=input, outputs=[output1, output2])

model.compile(loss={"label": tf.keras.losses.CategoricalCrossentropy(from_logits=False),"coordinates": tf.keras.losses.MeanSquaredError()},
              optimizer='adam', metrics=['accuracy'])

#model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
#              optimizer='adam', metrics=['accuracy'])

#Let's create a simulated data:
x_train=tf.random.uniform((55,75,75,1))
y_train_label=tf.random.uniform((55,10))
y_train_coordinate=tf.random.uniform((55,4))
y_train=[y_train_label,y_train_coordinate]

model.fit(x_train,{"label":y_train_label, "coordinates":y_train_coordinate},batch_size=120, epochs=5, verbose=1)

#model.fit(train, batch_size=120, epochs=5, verbose=1)

...結果你應該得到類似的東西：

Answer 2

在我的情況下，我遇到了同樣的錯誤（“logits 和標簽必須具有相同的第一維，得到 logits 形狀 [100,2] 和標簽形狀 [300]”），我做了 2 處更改：

1.- 我將“SparseCategoricalCrossentropy”更改為“CategoricalCrossentropy”

2.- 我將“tf.keras.layers.Dense(2)”更改為“tf.keras.layers.Dense(3)”

基本上我認為我的錯誤是因為我試圖向 model 引入 3 個輸入，因為最初，model 僅用於 2 個參數（我剛剛開始使用 AI）。

import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import os
import numpy as np
import matplotlib.pyplot as plt


carpeta_entrenamiento = './Casos_visiónGráficos/training'
carpeta_validacion = './Casos_visiónGráficos/test'

carp_entren_gatos = './Casos_visiónGráficos/training/cat200x200'  # imagenes de gatos para pruebas
carpeta_entren_perros = './Casos_visiónGráficos/training/wild200x200'# imagenes de zorr para pruebas
carpeta_entren_caballos = './Casos_visiónGráficos/training/horse200x200'
carpeta_val_gatos = './Casos_visiónGráficos/test/catTest'  # imagenes de gatos para validacion
carpeta_val_perros = './Casos_visiónGráficos/test/wildTest' # imagenes de zorr para validacion
carpeta_val_caballos = './Casos_visiónGráficos/test/horseTest'

#Guardar el numero de datos de entrenamiento para cada cosa
num_gatos_entren = len(os.listdir(carp_entren_gatos))
num_perros_entren = len(os.listdir(carpeta_entren_perros))
num_caballos_entren = len(os.listdir(carpeta_entren_caballos))

num_gatos_val = len(os.listdir(carpeta_val_gatos))
num_perros_val = len(os.listdir(carpeta_val_perros))
num_caballos_val = len(os.listdir(carpeta_val_caballos))

total_entrenamiento = num_gatos_entren + num_perros_entren + num_caballos_entren
total_val = num_gatos_val + num_perros_val + num_caballos_val

TAMANO_LOTE = 100
TAMANO_IMG = 200

#Aumento de datos (escala, rotacion, blabla)
print("Realizando aumento de datos")
image_gen_entrenamiento = ImageDataGenerator(
      rescale=1./255,
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')

#Generacion de datos de entrenamiento FTW
data_gen_entrenamiento = image_gen_entrenamiento.flow_from_directory(batch_size=TAMANO_LOTE,
                                                     directory=carpeta_entrenamiento,
                                                     shuffle=True,
                                                     target_size=(TAMANO_IMG,TAMANO_IMG),
                                                     class_mode='categorical')

#Generacion de datos de validacion
image_gen_val = ImageDataGenerator(rescale=1./255)

data_gen_validacion = image_gen_val.flow_from_directory(batch_size=TAMANO_LOTE,
                                                 directory=carpeta_validacion,
                                                 target_size=(TAMANO_IMG, TAMANO_IMG),
                                                 class_mode='categorical')

#Modelo!
modelo = tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(32, (3,3), activation='relu', input_shape=(200, 200, 3)),
    tf.keras.layers.MaxPooling2D(2, 2),

    tf.keras.layers.Conv2D(64, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),

    tf.keras.layers.Conv2D(128, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),

    tf.keras.layers.Conv2D(128, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),

    tf.keras.layers.Dropout(0.5),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dense(3)
])

#Compilación
modelo.compile(optimizer='adam',
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),#se remplazó para 3 entradas SparseCategoricalCrossentropy
              metrics=['accuracy'])


print("Entrenando modelo...");
epocas=60
history = modelo.fit_generator(
    data_gen_entrenamiento,
    steps_per_epoch=int(np.ceil(total_entrenamiento / float(TAMANO_LOTE))),
    epochs=epocas,
    validation_data=data_gen_validacion,
    validation_steps=int(np.ceil(total_val / float(TAMANO_LOTE)))
)

print("Modelo entrenado!");