search

honeygupta / PAGSR

This repository contains the code for our paper: Pyramidal edge-maps and Attention based Guided thermal image Super-Resolution (PAGSR)
11 stars 0 forks source link

KAIST dataset(5581) used in this work #4

Open YuanPeitao opened 2 years ago

YuanPeitao commented 2 years ago

Hello, author. We can't know which part (5581) of KAIST dataset (95328) you used from the currently disclosed resources. I hope you can share your KAIST dataset (5581) or cleaning method. Thank you very much for your work and look forward to your reply.

honeygupta commented 2 years ago

Hi. We sampled the images randomly and unfortunately we do not have the file list available with us. I understand that this is not ideal but given the nature of the experiment that we performed, I believe a uniformly sampled split should provide a similar behavior for your experiments

YuanPeitao commented 2 years ago

Hi. We sampled the images randomly and unfortunately we do not have the file list available with us. I understand that this is not ideal but given the nature of the experiment that we performed, I believe a uniformly sampled split should provide a similar behavior for your experiments

Thank you for your answer.

We still have a few questions to ask you: First, KAIST dataset is different from other datasets. We want to confirm whether you use single channel image calculation method or three channels to calculate PSNR in your work. In our experiment, we found that the index of KAIST testset is quite different from that in your paper. Secondly, can you share the FLIR dataset you aligned for us? We also conducted the same manual alignment, but for the fairness of experimental comparison, we want to use the dataset you used in the paper. If you can, please contact me by email: yuanpeitao2020@163.com .

Thank you for your previous work, and we will thank you in our results.

honeygupta commented 2 years ago

We use only one channel across datasets to compute the PSNR as thermal images are single channeled. We would be happy to share the aligned dataset but I believe we do not have the rights to share. You can find few samples from the dataset in our more recent work's repo - https://github.com/honeygupta/UGSR.

YuanPeitao commented 2 years ago

We use only one channel across datasets to compute the PSNR as thermal images are single channeled. We would be happy to share the aligned dataset but I believe we do not have the rights to share. You can find few samples from the dataset in our more recent work's repo - https://github.com/honeygupta/UGSR.

Thank you for your answer.

We will adjust the dataset and try again according to the sample images.

Can you share with us the original image of visual effect comparison in the paper?

Thank you again for your sincere reply and best wishes.