jaewon-lee-b / ltew Goto Github PK

Hi, Jaewon Lee:
I would like to ask how to generate a dataset with transformations for testing? How are the corresponding test set images and the corresponding forward transformation matrices generated? Can you provide the code?

Hi, Jaewon Lee:
I would like to ask how to generate a dataset with transformations for testing? How are the corresponding test set images and the corresponding forward transformation matrices generated? Can you provide the code?

Hello, the isc dataset link of SET14 is the same as that of osc, can you update it?

Hi, Jaewon Lee:

I used the pre-trained LTEW-RCAN for asymmetric SR, but I did not find the benchmark datasets for this task. Could you please kindly provide the benchmark datasets for asymmetric SR? or Could you please provide codes for generating testing benchmark datasets?

Hello, thanks very much for your intelligent work about image super-resolution (LTE) and image warping (LTEW), which gives me much guidance.

My problem is that: when I use LTE or LIIF for training the 360 image super-resolution task, using datasets provided by LAU-Net [1], the ws-psnr metric degrades during training, and the performance is bad. However, when using EDSR with pixel-shuffle for up-sampling, the training process has no problem.

I wonder if you have encountered similar problems, and if you have some possible solutions to address training on 360 images (ERP images).

[1] LAU-Net: Latitude Adaptive Upscaling Network for Omnidirectional Image Super-Resolution

Hello, the isc dataset link of SET14 is the same as the osc of SET5, can you update it?