你好，我正在尝试构建一个模型，该模型将从低分辨率和高分辨率图像的输入输出超分辨率图像，我的第一个错误是：UserWarning: Discrepancy between trainable weights and collected trainable weights, did you set
without callingmodel.trainable之后？，首先我尝试以这种方式解决，UserWarning: Discrepancy between trainable weights and collected trainable weights error(UserWarning: Discrepancy between trainable weights and collected trainable weights error) 但它没有用，因为 ISR 不显然没有“编译”属性。阅读完文档后，我确实认为我明白了，但这次我遇到了这个错误，我知道不可迭代一般意味着什么，我只是不明白这里有什么关系。model.compile
import os
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.models import Model
from ISR.models import RRDN, Discriminator
from ISR.models import Cut_VGG19
from ISR.train import Trainer
# Define paths for HR and LR input images
lr_train_dir = 'C:/data/lr_train_150/',
hr_train_dir = 'C:/data/hr_train_150/',
lr_valid_dir = 'C:/data/lr_test_150/',
hr_valid_dir = 'C:/data/hr_test_150/',
lr_train_patch_size = 22 #size of my LR image
layers_to_extract = [5, 9]
scale = 10
hr_train_patch_size = lr_train_patch_size * scale # 220 Size of my HR image
# Instantiate models
rrdn = RRDN(arch_params={'C':4, 'D':3, 'G':64, 'G0':64, 'T':10, 'x':scale}, patch_size=lr_train_patch_size)
discriminator = Cut_VGG19(patch_size=hr_train_patch_size, layers_to_extract=layers_to_extract)
feature_extractor = Discriminator(patch_size=hr_train_patch_size, kernel_size=3)
# Define optimizer and loss function
optimizer = Adam(1e-4, beta_1=0.9, beta_2=0.999)
#loss = 'mse'
loss_weights = {
'generator': 0.0,
'feature_extractor': 0.0833,
'discriminator': 0.01
}
losses = {
'generator': 'mae',
'feature_extractor': 'mse',
'discriminator': 'binary_crossentropy'
}
learning_rate = {'initial_value': 0.0004, 'decay_factor': 0.5, 'decay_frequency': 30}
log_dirs = {'logs': './logs', 'weights': './weights'}
flatness = {'min': 0.0, 'max': 0.15, 'increase': 0.01, 'increase_frequency': 5}
# Define trainer
trainer = Trainer(generator=rrdn,
discriminator=discriminator,
feature_extractor=feature_extractor,
#name='srgan',
log_dirs=log_dirs,
#checkpoint_dir='./models',
learning_rate=learning_rate,
losses=losses,
flatness = flatness,
loss_weights = loss_weights,
adam_optimizer=optimizer,
lr_train_dir = 'C:/data/lr_train_150/',
hr_train_dir = 'C:/data/hr_train_150/',
lr_valid_dir = 'C:/data/lr_test_150/',
hr_valid_dir = 'C:/data/hr_test_150/'
)
# Train the model
trainer.train(batch_size=16,
steps_per_epoch=20,
#validation_steps=10,
epochs=1,
#print_frequency=100
monitored_metrics={'val_generator_PSNR_Y': 'max'}
)
我得到的错误：
TypeError Traceback (most recent call last)
Cell In[18], line 52
46 flatness = {'min': 0.0, 'max': 0.15, 'increase': 0.01, 'increase_frequency': 5}
47 # Define feature extractor
48 #vgg = Model(inputs=rrdn.input, outputs=rrdn.get_layer('features').output)
49 #vgg.trainable = False
50
51 # Define trainer
---> 52 trainer = Trainer(generator=rrdn,
53 discriminator=discriminator,
54 feature_extractor=feature_extractor,
55 #name='srgan',
56 log_dirs=log_dirs,
57 #checkpoint_dir='./models',
58 learning_rate=learning_rate,
59 losses=losses,
60 flatness = flatness,
61 loss_weights = loss_weights,
62 adam_optimizer=optimizer,
63 lr_train_dir = 'C:/data/lr_train_150/',
64 hr_train_dir = 'C:/data/hr_train_150/',
65 lr_valid_dir = 'C:/data/lr_test_150/',
66 hr_valid_dir = 'C:/data/hr_test_150/'
67 )
69 # Train the model
70 trainer.train(train_lr_dir=lr_train_dir, train_hr_dir=hr_train_dir,
71 valid_lr_dir=lr_valid_dir, valid_hr_dir=hr_valid_dir,
72 batch_size=16,
(...)
77 monitored_metrics={'val_generator_PSNR_Y': 'max'}
78 )
File ~\anaconda3\envs\img_tf\lib\site-packages\ISR\train\trainer.py:104, in Trainer.__init__(self, generator, discriminator, feature_extractor, lr_train_dir, hr_train_dir, lr_valid_dir, hr_valid_dir, loss_weights, log_dirs, fallback_save_every_n_epochs, dataname, weights_generator, weights_discriminator, n_validation, flatness, learning_rate, adam_optimizer, losses, metrics)
102 elif self.metrics['generator'] == 'PSNR':
103 self.metrics['generator'] = PSNR
--> 104 self._parameters_sanity_check()
105 self.model = self._combine_networks()
107 self.settings = {}
File ~\anaconda3\envs\img_tf\lib\site-packages\ISR\train\trainer.py:163, in Trainer._parameters_sanity_check(self)
151 check_parameter_keys(
152 self.learning_rate,
153 needed_keys=['initial_value'],
154 optional_keys=['decay_factor', 'decay_frequency'],
155 default_value=None,
156 )
157 check_parameter_keys(
158 self.flatness,
159 needed_keys=[],
160 optional_keys=['min', 'increase_frequency', 'increase', 'max'],
161 default_value=0.0,
162 )
--> 163 check_parameter_keys(
164 self.adam_optimizer,
165 needed_keys=['beta1', 'beta2'],
166 optional_keys=['epsilon'],
167 default_value=None,
168 )
169 check_parameter_keys(self.log_dirs, needed_keys=['logs', 'weights'])
File ~\anaconda3\envs\img_tf\lib\site-packages\ISR\utils\utils.py:45, in check_parameter_keys(parameter, needed_keys, optional_keys, default_value)
43 if needed_keys:
44 for key in needed_keys:
---> 45 if key not in parameter:
46 logger.error('{p} is missing key {k}'.format(p=parameter, k=key))
47 raise
TypeError: argument of type 'Adam' is not iterable
ISR 文档 协作上的基本 ISR 实现 当我在协作上尝试此操作时，我得到了相同的警告：用户警告：可训练权重与收集的可训练权重之间存在差异，您是否设置了
model.trainablemodel.compile