vkdt devel diary

GLLM · December 15, 2021, 2:10pm

Hi @hanatos

May I propose 2 ideas:

double click on the pic in the retouch module could get back to the gallery module, the same way the opposite work: double clicking on a gallery pic opens the retouch module
have a before / after split, to better see the effect of what you are doing

Thanks
GLLM

hanatos · December 15, 2021, 2:30pm

heya,

no opinion on that. i think the ‘e’ shortcut works the same both ways, and it’s ‘x’ on gamepad to enter and ‘o’ on gamepad to leave, so it’s not perfectly symmetric.
there’s the ab module for that. of course there’s no convenient/simple way to wire it from the gui (other than excessive clicking). this is on my list of streamlining-gui-things.

GLLM · December 15, 2021, 5:07pm

Thanks @hanatos
I wasn’t aware of any of the 2 solutions. Case closed for these 2

Is there any way to completely reset the pipeline config … I ususally break it 99% of time I play with vkdt, and my only solution is to close vkdt and delete the cfg, even tho the correction I did were OK !

Thanks

betazoid · December 15, 2021, 5:18pm

There is a reset-button in lighttable mode, I think under selected images. I already asked for a shortcut for this, it should be some complicated shortcut with 3 keys or so.

GLLM · December 15, 2021, 5:37pm

Thanks @betazoid, it works indeed. I didnt make the link between this and the pipeline config

da-phil · December 15, 2021, 10:16pm

@hanatos
I’m really excited about the the amount of possibilities this computation DAG enables us in vkdt, hence I wanted to try out how a DAG with multiple input sources (alignment workflow) would work.
After quickly skimming through your JCGT paper (really cool!) I’ve become curious and wanted to tinker around with the align module and seem to have not understood it yet.
I wanted to see the align outcome from 2 handheld frames which I created from a exposure bracketing set.
This is one of the frames:

And this is the very funky outcome:

Using this config: align2.cfg.txt (3.0 KB)

Can you shed some light on what I’m doing wrong here please?

hanatos · December 16, 2021, 10:09am

haha brilliant. may have to do with the blurs too. to match a region around a pixel, i’m blurring a difference image. also i found a fairly embarrassing bug in the motion reconstruction yesterday. if you could (privately?) share the source images too i can have a look at exactly your case. but i’m suspecting i’ll need to fix something in general first.

hanatos · December 16, 2021, 11:09am

(never mind i can reproduce here and i think i also fixed it with a fixed blur. unfortunately that one breaks other functionality, so i’ll hold back pushing for a bit… sorry)

hanatos · December 16, 2021, 3:05pm

this should now be fixed in git master.

hanatos · December 16, 2021, 3:06pm

this should now be fixed in git master. do you still have a checkout you can easily build to verify? i didn’t push an obs build yet because it was a rather invasive/intricate change, blurs are everywhere! on the other hand it probably doesn’t get more broken than now…

hanatos · December 16, 2021, 3:09pm

…should be fixed too (but also only in git currently)

da-phil · December 16, 2021, 3:26pm

I just pulled, but now the results only looks slightly different

With another pair of images, which has almost no difference in exposure it worked better, but there are still plenty of artifacts visible.
I’ll PM you the images, then you can try to reproduce.

betazoid · December 16, 2021, 5:12pm

Thanks. I will check soon. I was also going to check if vkdt can be compiled on Ubuntu 20.04 with Vulkan sdk + Nvidia.

betazoid · December 17, 2021, 10:40am

I just compiled vkdt on Manjaro and tested the zones module. Something is still wrong with it, I think it is something with the gamma slider. But it is hard to describe for me what it is. I tested it with two different photos.
I will upload the raw of the candle pic so maybe I can explain it better.

IMG_20211211_175435.dng (30.6 MB)

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Here is a screenshot. The important thing is the visualisation of the zones module mask:

I would like the area of the candle to be a brighter grey or even white. But basically, no matter where I pull the gamma and the epsilon slider, apparently, I cannot change the shades of grey. So, if I make the darkest zones darker, the brighter zone where the candle is also gets darker. This used to work well before the fix - except that the picture changed when I went to lighttable mode and/or exported. Something seems to be very broken now.
The area around the flame always has the same middle grey colour, no matter how I change the sliders.

I also tested this with a playraw from here.

Here too, apparently it is impossible to change the shades of grey. I would like to make the zone where the foreground is much darker (in the zones visualisation), maybe even completely black but that’s not possible.

Or I don’t know, maybe it’s supposed to be like this? I think it did work well before the fix in the preview.

hanatos · December 17, 2021, 11:14am

oops, please try again:

thanks for quickly spotting these things. i had cleaned up the api of the guided filter and wired input and guide image the wrong way around.

betazoid · December 17, 2021, 11:37am

seems to be functional now

phweyland · December 17, 2021, 11:37am

Thanks. Trying to build vkdt again, but it may take time. I’ve again some trouble with openmp lib:

git/vkdt/bin/modules/i-raw/libi-raw.so: undefined symbol: omp_set_max_active_levels

while the debian distrib finds the library (and works).
I’ve restarted from a clean vkdt clone, submodule init & update, …

hanatos · December 17, 2021, 11:40am

…pushed obs packages now.

giannissc · December 17, 2021, 2:03pm

Is there a chance that vkdt will be released as a snap/flatpak/appimage to make it easier to test?

darix · December 17, 2021, 2:07pm

the native distro packages do not work for you?