A few missing lines in "generate_images_using_image_cap_dataset.py"

[JooyoungLeeETRI]

Dear authors,

Thank you very much for sharing your valuable source codes. I've found a few missing lines while running your code on my side, as follows:

In "generate_images_using_image_cap_dataset.py",

... from datasets_image_cap import Image_Cap_pair_dataset ... def main(argv): ... psnr = compute_psnr(x, x_hat) try: ms_ssim = ms_ssim_func(x, x_hat, data_range=1.).item() except: ms_ssim = ms_ssim_func(torchvision.transforms.Resize(256)(x), torchvision.transforms.Resize(256)(x_hat), data_range=1.).item() lpips_score = loss_fn_alex(x, x_hat).item() ...

Additionally, I found a couple of bugs as follows:

• There's no "ms_ssim_func" function declared.
• There's no "image_list" variable.

Please note that I've just tried to run your code over the Kodak image set.

[hagyeonglee]

Hi, Jooyoung.

Thank you for being so interested in our project :)

1. To measure MS-SSIM metrics first, install it via PyPI and declare the function at the top:

pip install pytorch-msssim

from pytorch_msssim import ms_ssim as ms_ssim_func

2. image_list is the path to the dataset you want to test.

You can find instructions on how to do this at Example LINK.

[minguinho26]

Hi Jooyoung.

Thank you for reporting the problem with the inference code. Also thanks to Hagyeong for answering quickly.

I fixed the inference code. If you also find any problems, please write the question to issue board.