Announcement

Collapse
No announcement yet.

Using Nvidia AI to deflick and add missing content by analyzing frames or adding warning if separate buckets are darker or differ greatly from surrounding buckets.

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Using Nvidia AI to deflick and add missing content by analyzing frames or adding warning if separate buckets are darker or differ greatly from surrounding buckets.

    Hello Adele is here.

    1 ------------------------------------

    Judging by demos found on YT Nvidia AI is quite potent and capable for some tasks and i am not sure if this can be done with it so it's just idea...

    Can you guys implement some deflick (adjusting light variation), missing buckets and textures restoring by using some algorithm. I saw Nvidia AI is able to replace eyes, hair on people, books on shelves etc. so i guess this isn't that far off.

    Type of usage application scenarios:
    Rendering sequence of images for animation and from whatever network reason one machine not getting HDRI map for IBL or any other texture for 100 frames... resulting in slightly darker image or buckets if you are using DR.

    Solution workflow:
    Can Nvidia AI or any other type of solution analyze - 5 and +5 frames and if those discrepancies are noticeable, reconstruct affected buckets later on without need to re-render entire sequence.

    I guess it can use light, shadow, noise and diffuse pass and see if texture is having 16x16 miss match between frames, i guess it is same how denoiser works.

    2 ------------------------------------

    Another closely related wishful thinking

    Type of usage application scenarios:
    Using 10 machines in DR with 16x16 buckets was never great when you need to find if something is a miss, causing darker bucket.
    It can be bounce from missing texture on object that renders fine on machine 1 and causing dark indirect light result on machine 2 missing textures as that second machine never rendered buckets machine 1 did you can never know what causing it (i know it's confusing...)

    This is big headache when there is 200-300 cores render 16x16 buckets as you cannot read names on DR Server Render element and colors are very similar sometimes... Resulting either in adjusting bucket size to 128x128 and fire machine by machine to determine what causing problems. Which can take ages.
    I know there is transfer missing assets, but that does not account that there are assets who might have same name, missing plugins, older/newer versions of software or even other network/windows issues...

    Solution workflow:
    Is it possible to implement algorithm which can analyze whole image and show warning in console "from 256 buckets rendered, 34 (machine name) buckets are substantially darker from surrounding buckets, disregard if everything is fine or check...
    You can either dismiss or act on this warning. It doesn't even have to do this all the time, i guess if GPU does this analysis every 500 buckets or whatever.

    I know this is hard to implement as image content can vary in between very much so i guess maybe adding this option as something that would need 2 separate images rendered and compare those 2 for slight variations in missing buckets.
    Which is again much faster than manually checking everything. as you can simply submit render on DR on machines that you know are fine, fix machine that causing issue and add it during rendering as that is already supported right now.

    Thanks and take care,
    Ivan










Working...
X