Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Training on greyscale dataset #243

Open
Dolabok opened this issue Jun 25, 2023 · 1 comment
Open

Training on greyscale dataset #243

Dolabok opened this issue Jun 25, 2023 · 1 comment

Comments

@Dolabok
Copy link

Dolabok commented Jun 25, 2023

Hi, is it possible to train the model on a grayscale dataset without transforming it to rgb? I tried the example code with c_dim=1 in the RRDN model but I have an error with the dimensions of the next layers:

WARNING:tensorflow:Model was constructed with shape (None, 80, 80, 3) for input KerasTensor(type_spec=TensorSpec(shape=(None, 80, 80, 3), dtype=tf.float32, name='input_2'), name='input_2', description="created by layer 'input_2'"), but it was called on an input with incompatible shape (None, 80, 80, 1).
Traceback (most recent call last):
File "/home/dolabok/Documents/[...]/super_resolution.py", line 76, in
trainer = Trainer(
File "/home/dolabok/anaconda3/lib/python3.9/site-packages/ISR-2.2.0-py3.9.egg/ISR/train/trainer.py", line 105, in init
File "/home/dolabok/anaconda3/lib/python3.9/site-packages/ISR-2.2.0-py3.9.egg/ISR/train/trainer.py", line 185, in _combine_networks
File "/home/dolabok/anaconda3/lib/python3.9/site-packages/keras/utils/traceback_utils.py", line 70, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/home/dolabok/anaconda3/lib/python3.9/site-packages/keras/engine/input_spec.py", line 277, in assert_input_compatibility
raise ValueError(
ValueError: Exception encountered when calling layer "discriminator" (type Functional).

Input 0 of layer "Conv_1" is incompatible with the layer: expected axis -1 of input shape to have value 3, but received input with shape (None, 80, 80, 1)

Call arguments received by layer "discriminator" (type Functional):
• inputs=tf.Tensor(shape=(None, 80, 80, 1), dtype=float32)
• training=False
• mask=None

it seems that the discriminator is hard-coded with channel = 3

My code :

# Initialize the models
lr_train_patch_size = 40
layers_to_extract = [5, 9]
scale = 2
hr_train_patch_size = lr_train_patch_size * scale

rrdn = RRDN(arch_params={'C':4, 'D':1, 'G':64, 'G0':64, 'T':10, 'x':scale}, c_dim=1, patch_size=lr_train_patch_size)
f_ext = Cut_VGG19(patch_size=hr_train_patch_size, layers_to_extract=layers_to_extract)
discr = Discriminator(patch_size=hr_train_patch_size, kernel_size=3)

# Training settings
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}
flatness = {'min': 0.0, 'max': 0.15, 'increase': 0.01, 'increase_frequency': 5}
log_dirs = {'logs': './logs', 'weights': './weights'}

# Initialize the trainer
trainer = Trainer(
    generator=rrdn,
    discriminator=discr,
    feature_extractor=f_ext,
    lr_train_dir='low_res_train',
    hr_train_dir='high_res_train',
    lr_valid_dir='low_res_train',
    hr_valid_dir='high_res_train',
    loss_weights=loss_weights,
    learning_rate=learning_rate,
    flatness=flatness,
    dataname='image_dataset',
    log_dirs=log_dirs,
    weights_generator=None,
    weights_discriminator=None,
    n_validation=40,
)
    
# Train the model
trainer.train(
    epochs=100,
    steps_per_epoch=500,
    batch_size=16,
    monitored_metrics={'val_PSNR_Y': 'max'}
)
@Dolabok
Copy link
Author

Dolabok commented Jun 26, 2023

you are using the VGG19, which cannot use grayscale images

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant