viewer.rs has a different render result with render.rs

[oUp2Uo]

Thanks for the great work, it is very useful to view the compressed npz model.

Today, occasionally, I found viewer.rs has a different render result with render.rs for a same model on the same camera position. And this is the first position.

viewer.rs

render.rs

Not all position different, just 1st in this image sequence(~60).
I have tried to change the rendering order of test and train, and this time the 1st in train got error.
It seems that render.rs's result is not correct at the first position.

Do you have any idea about this?
Thank you.

[KeKsBoTer]

Nice reconstruction!
This seems to be an issue with the sorting. I have to check if something is incorrect with the setup there :/

[oUp2Uo]

Thanks for the reply.
I have done some more test, and edited the article.
It seems that only the 1st image has error.
Maybe some bug in initialization?

[oUp2Uo]

Hi, after some more tests, I found other differences between video.rs and render.rs:
render.rs could output images in any resolution, but video.rs could not.
For example, 4k x 2k may ok, but 4001 x 2001 could not.

I found function download_texture is different in these two program.
And wgpu::TextureFormat is also different.

[oUp2Uo]

Hi, with these days' tests, I found this bug not only exist in render.rs, but also video.rs.
Although they have different download_texture and download_buffer functions, there would be error render result image in these two program.