Yi Wang

All experiments in the paper are conducted with rectangle masks. cmyyy 于2019年7月20日周六 下午3:46写道： > Another question,in the quanttative experiment, the places2 model was > trained on random strokes or rectangle...

https://github.com/shepnerd/inpainting_gmcnn/issues/20#issuecomment-506572261

Need more detailed information (e.g., error descriptions) for reference.

You mentioned pretrained models are for tensorflow instead of pytorch. You can try it with tensorflow testing code.

For your other two questions, 2) It is hard to determine the optimal training images. The basic rule is, the required training number is relatively small when the data are...

The pretraining stage of using rect masks differs a lot from that using stroke masks (the former needs to predict more areas than the later), so it will cause such...

You can try it again after commenting L13-15 in test.py out. The mentioned error should be caused by these lines (https://github.com/shepnerd/outpainting_srn/blob/49be8beca8571a50788f9f1b392e9ec4e6de13fd/test.py#L13), which are used to select a GPU for testing...

The padding size calculation using dilation conv is a bit different from the common conv, and you highlighted part just hard-codes these parameters. For the detailed rules, you can refer...

Using the option ``--load_model_dir [Your model path]`` to continue the training. The option ``--pretrain_network 0|1`` is to decide whether using the all training losses or just the reconstruction loss.

> > It happens as the loss for updating generator in WGAN-GP could be negative. > thank you very much. I have another question for you, why G_ Does loss...