MacGruber_Laboratory
patreon
Monthly Report: July
🕑 Added 2020-07-27 19:33:35 +0000 UTCTime for another monthly report! This time there is also a poll at the end, I would appreciate if you could take the time to answer it!
I'm posting a report like this near the end of each month, reporting on what I worked on behind the scenes and what the vague plans for the next month are. It is intended to help you decide if you want buy-in on the coming month or not.
IdlePoser
This month I released another IdlePoser update, focusing on debug rendering and UI, trying to make it easier to use. This was in response to feedback that the plugin was too complicated to understand. Well, making something complex easy to use is very hard.
Anyway behind the scenes I also worked on some other planned features, which did not yet result in anything releasable, yet:
- Arc-Length-Parameterization: This is a feature intended to improve animation quality as it normalizes the velocity while blending between poses. On top of the normalized velocity I can then add a proper SmoothStep for ease in/out. However, after basically finishing the implementation and already working on my demo scene, I realized that this makes only sense for either positional or rotational changes. You can only use one or the other for normalizing velocity, not both. If the wrong method is chosen, it can actually make animation a lot worse. Technically one would have to configure which method to chose for every transition for every controller. That's exactly the kind of combinatoric explosion I'm trying to avoid with IdlePoser. Not sure how to solve this yet.
- Vertex Anchoring system: Also did some research on the planned vertex anchoring. The idea here is that instead of picking a bone node to anchor against, you could pick an actual vertex. In theory this could allow to anchor with a lot more precision, since it could be directly where for example a hand touches the body, not several centimeters away. Also it would respect morphs, meaning it could adjust to different Looks more easily, say small vs. large breast. By now I managed to access the morphed mesh in T-Pose before skinning, meaning the positions of the control nodes have not yet been applied. So I have to figure out how to clone the exact way of VaM doing skinning on the GPU on my own while doing it for a single vertex on the CPU. Not impossible, just takes time.
Since I have been working on IdlePoser for more or less 3 months straight, I'm gonna switch to other things for a while. No worries, I will come back to IdlePoser eventually.
SuperResolution for Video?
In the last few days I have been working on some ideas for improving my SuperResolution plugin (which is part of the Essentials package). The idea is to implement GPU image downscaling already in VaM, which has two benefits:
- Saving a screenshot would be way faster. Taking the screenshot itself doesn't take long, only saving the file takes ages, so no more 1-2 second freezes when taking screenshots. (If you actually want those 8K images and downscale yourself, you will still have the option, of course.)
- Once we have GPU downscaling, I could also hook SuperResolution to other cameras in the game. E.g. the WindowCamera or the regular desktop player camera. Meaning your video recordings can benefit from better AA as well as better hair and shadow quality you get from using SuperResolution. If you have the GPU power, in theory you could also just activate it for general desktop mode. (In VR you already got the RenderScale slider)
Right now I have a simple linear downscale already implemented. However, that's of course not optimal quality, as it makes the image slightly blurry. Ideally I should probably implement Lanczos resampling, which is what most software uses these days. However, all the explanations and shader implementations I found so far are only for upscaling, which is easier, but we need downscaling.
Here is a comparison of the methods at 200% zoom. Image is also attached to the post as download, so you can look at it without Patreon messing with it. First image is native VaM, middle image where I'm at right now, last image is where I'm trying to get to. Look at hair vs. background and at the eyes.
Poll: Are these MonthlyReports interesting?
Anyone actually reading my walls of text? I'm trying to figure out if it actually makes sense writing these or should I rather spend time of making useful stuff? I have some 400+ followers right now, so if only 5 people are actually interested in reading this, it's probably not worth spending time on ;)
