same as the title. I have tried all the models but the result is same. I am uploading some pictures to get the idea

Currently only one GPU can be used. Can multiple Gpus be trained at the same time? One GPU is too slow.

Hello,

In your test.ipynb, you create a Generator with cnum_in=5, but you pass as input an image with 3 channels only.

Is it up-to-date with the current train.py code ? Should you also concatenate the ones and ones*mask to the image channel to pass as input, like in the training ?

Thanks.

If you read Readme, you can download it through the link below, but there is no pth file if you download and unzip it.

https://drive.google.com/u/0/uc?id=1tvdQRmkphJK7FYveNAKSMWC6K09hJoyt&export=download

Hello!

First I want you to thank you because you did a really great job migrating this awesome DeepFillV2 to Python.
Also than you for implementing the GUI. @nipponjo

But I cannot make this work when converting it to Core ML.
Is it nearly done but when is about to finish I get this error:

Converting PyTorch Frontend ==> MIL Ops: 84%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉ | 518/619 [00:00<00:00, 3312.17 ops/s]
Error during Core ML conversion: ('Op "706" (op_type: conv_transpose) Input weight must be const at compile time', 'weight', 'wi_center')

Do you have any idea why this is happening?
Thanks in advance

Thank for your work. And I'm getting an error that I can't solve. I would like to ask you how to solve it and do you have relevant experience.
plea

just like this？

Hello,

I trained the model and images generated looked nice, but during inference, it's pretty bad. I even try to infer using training data (just to test), and even then, it's pretty bad, so I'm debugging it, making sure images at inference and training get the same preprocessing.

I noticed that you used a grid during the Generator inference, but not during training, why is that ?

Thank you.

你好，感谢你的分享，想咨询一下可以说一下你当前环境的版本吗 特别是torch版本，我这边不知道是版本高了还是低了，报错

It cannot run....

Is there a straightforward way to adapt the model to grayscales (one channel) images?

Hi
Thanks for the repo!
I am running your provided training configuration on the celeba dataset, with 128x128 images. I get the following results in tensorboard:

Is this expected?

当我把代码下载下来并且运行app.py后发现报错，似乎是缺少了某个文件

Can you please create new CoreML model ？

What data set was used to train the places and celebahq model?

Hello,Thanks for your code! I would like to know how do you generate your test masks? I find that when i try to use pretrained models to test on my masked images,the results seems to be not very good.

What do you mean by the value below that?
For 1280x720 images, I would like to ask you how to set the values below.

mask options

height: 256
width: 256
max_delta_height: 32
max_delta_width: 32
vertical_margin: 0
horizontal_margin: 0

Thanks for the nice code! Could you given an estimate about the training time of the model on CeleBA-HQ and Places2 ? And how is the model's performances with a smaller dataset with, for instance, thousands of data?

Hello @nipponjo , I've exported your work to onnx but it didn't work
This is script

import torch

pretrained = "pretrained/states_pt_places2.pth"

generator_state_dict = torch.load(pretrained)['G']

if 'stage1.conv1.conv.weight' in generator_state_dict.keys():
    from model.networks import Generator
else:
    from model.networks_tf import Generator  

# set up network
generator = Generator(cnum_in=5, cnum=48, return_flow=False)

generator.load_state_dict(generator_state_dict, strict=True)

img = torch.rand([5, 512, 512]).unsqueeze(0).cpu()
mask = torch.empty(img.shape[0], 1, img.shape[2], img.shape[3]).cpu()
data = (img, mask)

generator.cpu().eval()

dynamic_axes = None
dynamic_export = True
if dynamic_export:
    dynamic_axes = {
        'input': {
            0: 'batch',
            2: 'height',
            3: 'width'
        },
        'output': {
            0: 'batch',
            2: 'height',
            3: 'width'
        }
    }
with torch.no_grad():
    torch.onnx.export(
        generator,
        data,
        "output.onnx",
        input_names=['input'],
        output_names=['output'],
        export_params=True,
        keep_initializers_as_inputs=False,
        verbose=False,
        opset_version=12,
        dynamic_axes=dynamic_axes)
print(f'Successfully exported ONNX model')

End the error was from ONNX

Traceback (most recent call last):
  File "deepfillv2-pytorch/export2onnx.py", line 65, in <module>
    torch.onnx.export(
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/__init__.py", line 316, in export
    return utils.export(model, args, f, export_params, verbose, training,
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/utils.py", line 107, in export
    _export(model, args, f, export_params, verbose, training, input_names, output_names,
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/utils.py", line 724, in _export
    _model_to_graph(model, args, verbose, input_names,
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/utils.py", line 497, in _model_to_graph
    graph = _optimize_graph(graph, operator_export_type,
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/utils.py", line 216, in _optimize_graph
    graph = torch._C._jit_pass_onnx(graph, operator_export_type)
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/__init__.py", line 373, in _run_symbolic_function
    return utils._run_symbolic_function(*args, **kwargs)
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/utils.py", line 1032, in _run_symbolic_function
    return symbolic_fn(g, *inputs, **attrs)
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/symbolic_helper.py", line 172, in wrapper
    return fn(g, *args, **kwargs)
  File "/miniconda3/envs/py39/lib/python3.9/site-packages/torch/onnx/symbolic_opset9.py", line 1281, in _convolution
    raise RuntimeError("Unsupported: ONNX export of convolution for kernel "
RuntimeError: Unsupported: ONNX export of convolution for kernel of unknown shape.

I've google it but they said that because there is an operation in pytorch that onnx doesn't support. But I could not investigate what operation.
Could you help me please? Thank you

How to set the train and value folder? The good pictures and the bad pictures are placed in one folder?

Hello.

I'm trying to learn an image with an image size (720, 1280, 3), and I've confirmed that the initial ether is 1,000,000.

Do we usually proceed with the learning until the iter reaches 1,000,000?
Currently (mine)
iter=180000,
d_loss=1.0000
g_loss=-0.2722
ae_loss=0.0576
ae_loss1=0.0150
ae_loss2=0.0425.

I wonder if it is right to converge to 1 for d_loss.