Weighted Avatar Performance Ranking | Voters

Weighted Avatar Performance Ranking

tracked

Maribel Hearn

A similar post was made, but was closed because the feature was still in beta at the time. This feature is now live. This feature request is also different from other requests in that I am requesting a weighted average instead of removal or change in values.
With the introduction of the avatar performance ranking system, it has become very clear that some extremely optimized avatars are being ranked as poor or worse because they are not optimized in one category alone. For example, a hyper-optimized avatar with, lets say 10,000 polys, one material, no cloth physics, one mesh, no particles or audio sources, but with 51 dynamic bone transforms. This avatar would be ranked poorly, despite being very well optimized. I propose a weighted average that takes the performance ranking of all of your performance stats into account, and gives you an overall performance ranking, instead of the extremely harsh ranking of the worst rank being the one that is shown to all users. Of course, users with absurd numbers of cloth polygons or dynamic bone colliders should still be ranked poor, even if they have a low poly count and are a single material. This is where the weighting of the calculation comes in.
Please consider implementing this. Seeing an avatar that you spent hours optimizing (Merging bones, atlasing, etc) get ranked terribly because you have a dynamic skirt is discouraging, and removes the incentive to optimize in general, if even your most optimized avatars will be ranked as poor anyway. Why atlas, decimate, merge bones, limit collider usage, etc, if your performance ranking is going to be "poor" even with all the optimizations?

December 21, 2018

MisoNyah

(dynamic) % scale would be nice.

Also, maybe collect performance data from clients as every device/PC/world might be different.

kawashirov

various realtime lights do different impact on performance.

TéaWithBiscuits

Think should be a point system based on amount bones, animator layers, states, any unity component based on the impact it has, combine a point and make a scale of how optimized it is.

Godfall

Sounds like a great idea, I would also like to see the polygon count limit slightly expanded for "Medium" and "Poor" ranking. It makes absolutely no sense that an avatar that would be "Good" becomes "Very Poor" from being a singular polygon over. vrc pls fix

Nіyah

Heartfully agree with this post, one main reason I am not even trying to optimize is the fact that having one bad ranking pull down the entire avatar and I just cba.

However, categories should have different weight in the calculation ( Materials being more bad than poly for examples, and so on)

初雪白子_WBmfmf

Based on my own and others' personal experience, people who don't use Blender often find it difficult to avoid exceeding the 70,000 polygon limit for their models. However, by using optimization tools, the number of SkinnedMeshRenderers and Materials tends to decrease. For example, VeryPoor, which was rated only for exceeding 70,000 polygons, is considered to have the same load as Medium, and we believe that by averaging or redefining it based on the weight of the evaluation, it will become an appropriate evaluation rank.

Scout - VRChat Head of Quality Assurance

marked this post as

tracked

DrBlackRat

A point system would probably be good for this, every component could add a different ammount of points (e.g. 1 Polygon = 1 Point, 1 Light = 400 Points, etc.) and depending on how many you have you get a rank.

ni1chigo2115

I think it is essential to rank the avatar performance considering the weight of setcalls and the texture size (there are few people who care about it unexpectedly).

If each person has more than 50MB of avatars in dozens of instances, they are all useless.

GotoFinal

https://vrchat.canny.io/feature-requests/p/improve-avatar-performance-rating-system-especially-for-cloth Also my improvment that is a bit related to this one.

→