jingjin25 / lfasr-geometry Goto Github PK

I have recently read your paper "Learning Light Field Angular Super-Resolution via a Geometry-Aware Network". When I click the link https://github.com/jingjin25/LFASR-geometry, there only readme doc. I am wondering if you could open the source again. Thank you.

Thank you for your work!
Can you share the HCI old dataset and Inria DLFD dataset?

Congratulations on your article being accepted by AAAI2020. I notice that the dataset using in this article is New HCI. But you didn't use training datasets and testing datasets strictly in accordance with their own partitions (e.g. Origami and dishes). Could you explain why you did that?

Thank you!

First of all, congratulations on your article being included by AAAI. The following error occurred when I reproduced your code.What code needs to be changed？

Hi Dr. Jin, thanks for sharing your great work. I tried your model on HCI and got the result as good as listed in your paper. But I cannot get the correct result on HCI-old. Could you please provide a demo on HCI-old? Thanks in advance!

I did not make any changes, however, I found the result is different as in paper

Did you use only the first four images in your bedroom for the test?

Hi, excellent work!
I test the pre-trained model "HCI_2x2-7x7.pth" on four LF data in HCI_new, including bedroom, bicycle, herbs, and dishes. But I got :
===> test
loaded model ./pretrained_model/HCI_2x2-7x7.pth
testing LF 0
psnr: 37.54, ssim: 0.948
testing LF 1
psnr: 30.84, ssim: 0.915
testing LF 2
psnr: 32.22, ssim: 0.867
testing LF 3
psnr: 28.56, ssim: 0.914
Comparing to your reported results in the paper, it is much lower performance. Do you have any idea what is wrong with the testing?

Looking forward to your reply!

我跑的时候爆显存