Error when running example code

[jcasey30]

I have been trying to get this repo to work but I am running into an issue. After cloning this repo and downloading the afhq dataset, I attemp to run the code using this command:

python main.py train --dataset_name afhq --dataset_path C:\Users\user\Documents\ATR\data\afhq --batch_size 4 --res 512 --config e --impl ref

Once training starts, however, I get an error returned:

` Traceback (most recent call last): File "main.py", line 99, in main() File "main.py", line 66, in main trainer.train() File "C:\Users\user\Documents\GitHub\StyleGAN2\train.py", line 150, in train G_loss, D_loss = self.train_step(real_images, real_labels) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\def_function.py", line 780, in call result = self._call(*args, **kwds) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\def_function.py", line 823, in _call self._initialize(args, kwds, add_initializers_to=initializers) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\def_function.py", line 696, in _initialize self._stateful_fn._get_concrete_function_internal_garbage_collected( # pylint: disable=protected-access File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\function.py", line 2855, in _get_concrete_function_internal_garbage_collected graph_function, _, _ = self._maybe_define_function(args, kwargs) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\function.py", line 3213, in _maybe_define_function graph_function = self._create_graph_function(args, kwargs) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\function.py", line 3065, in _create_graph_function func_graph_module.func_graph_from_py_func( File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\framework\func_graph.py", line 986, in func_graph_from_py_func func_outputs = python_func(*func_args, **func_kwargs) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\def_function.py", line 600, in wrapped_fn return weak_wrapped_fn().wrapped(*args, **kwds) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\eager\function.py", line 3735, in bound_method_wrapper return wrapped_fn(*args, **kwargs) File "C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\framework\func_graph.py", line 973, in wrapper raise e.ag_error_metadata.to_exception(e) ValueError: in user code:

C:\Users\user\Documents\GitHub\StyleGAN2\train.py:209 train_step  *
    G_loss, G_reg = self.loss_func.get_G_loss(real_images, real_labels, compute_reg=True)
C:\Users\user\Documents\GitHub\StyleGAN2\modules\losses.py:71 get_G_loss  *
    pl_lengths = tf.math.sqrt(tf.reduce_mean(tf.reduce_sum(tf.square(pl_grads), axis=2), axis=1))
C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\ops\gen_math_ops.py:10321 square  **
    _, _, _op, _outputs = _op_def_library._apply_op_helper(
C:\Users\user\AppData\Local\Continuum\anaconda3\envs\capsule\lib\site-packages\tensorflow\python\framework\op_def_library.py:486 _apply_op_helper
    raise ValueError(

ValueError: Tried to convert 'x' to a tensor and failed. Error: None values not supported.

`

[ammar-deep]

While I am unsure of the exact cause of your error, I will recommend that you ensure that your model's training images are properly loaded. I just cloned this repository and began training it on my custom dataset using all default settings except the --impl=cuda option, and everything appears to be working fine.

[ammar-deep]

So the error you mentioned is coming due to the loss ns_pathreg_r1() used for afhq and ffhq. Change the loss to ns_DiffAugment_r1() which doesn't need labels and it will work fine. Due to some unknown reason the number of labels are none when the model starts training while using the ns_pathreg_r1() loss.

[seymayucer]

I have tried many things to use GradientTape.gradient() but could not do it, I switch it to tf.gradients to use StyleGan2 ppl loss, it worked;

#from
       pl_grads = pl_tape.gradient(pl_noise_applied, pl_w)
#to
       pl_grads = tf.gradients(
           ys=pl_noise_applied,
           xs=pl_w
       )[0]