Digital-Colour-Alchemy-Club / image_formation_whatever

i said WHATEVER
GNU General Public License v3.0
1 stars 1 forks source link

Test patterns and images -- curation and workflow #2

Open zachlewis opened 3 years ago

zachlewis commented 3 years ago

Currently, we're using a 4K Marcie EXR as a test image. Marcie is great, but we need a more diverse set of imagery for properly evaluating rendering performance. (Right now, we also have Alex Fry's CLFTestImage v6, and slightly different Marcie encoding).

We discussed adding simple hue rainbow ramps, and a CC24 (for which we can use colour, obviously). What are you guys thinking? I have a few images in mind off the top of my head, but not sure about licensing. We should check in with ACES to see how things are going with their own licensing efforts...

Additionally, slinging 4K RGB arrays around isn't inexpensive -- computation overhead aside, it turns out cached functions will pretty quickly exhaust instance resources, crashing the instance, requiring a manual reboot.

We briefly discussed, and agreed, that working with resized proxies makes much more sense. I can't remember what we agreed to, in terms of a policy for scaling stuff, but perhaps as a quick fix, we can use simple bicubic downsampling to fit either the width (or height) of the original image to a fixed scale. Or maybe we can center crop.

I've started putting together a simple RemoteData class which handles downloading, caching, and retrieving local paths for data of a certain size. Perhaps we can formalize a LocalData class for computing and caching proxies for the curated test image library and user-uploaded images alike...

sobotka commented 3 years ago

Smooth sweeps are extremely useful here, across a range of dynamic levels to spot kinks in transitions both chromaticity speaking and luminance.

I'd think it would be relatively easy to snap together a Python snippet to generate an achromatic sweep from upper left where EV are 0.25 increments left to right, and, 0.25 increments top to bottom? Essentially a matrix of exposures for the achromatic axis?

The hue stripe in the Alex Fry repo is extremely useful as well, given it shows and isolates chromaticity skews extremely effectively.