Yi Wang

The mentioned train.txt is a file storing the full paths of the training images, like: `/data/face/0x1.png` `/data/face/0x2.png` `...` `/data/face/0f4.png` To generate the corresponding training file of a given dataset, you...

It seems some input options are invalid. The input for the option `--dataset` is just a dataset name like 'paris_streetview' instead of a path. If given a path, the creation...

About the experimental settings, you can refer to our paper. We use randomly cropped 256x512 (hxw) images from paris-streetview for training and 256x256 images in eval_gt for evaluation. And we...

We will release some of our other pretrained models soon.

It seems like this issue is caused by the inconsistent convolution filter number of the generator in your trained model and tested one. Please ensure using the same `--g_cnum` parameter...

You can pad 256x256 paris-streetvew images into 256x512 ones and then test them.

I guess so. The custom modules are only used in training, and the generator for testing is formed by standard modules in tensorflow, like convolutional layers.

In our paper, we used rectangle masks in random locations for testing. The released code and some pretrained models support the evaluation on the stroke masks. cmyyy 于2019年7月18日周四 上午10:55写道： >...

Yes. cmyyy 于2019年7月18日周四 下午1:19写道： > So one 128*128 rectangle mask at random location of an image, and it is > same for both the datasets,right? > > — > You...

Yes. 2k images are randomly chosen for evaluation since the full testing set of places2 is huge. About the shape problem, L63-70 in test.py just crop and resize the central...