Principal graphics architect @Arm. Previously @Unity3d, @PlayStation, and @Qualcomm. My views do not represent any employer. โต๐๐โท๐โฝ๐บ๐ป๐ฑ๐น๏ธ๐๏ธ๐๐จ๐ฆ๐ฌ๐ง
@Lazer@flysas In Denmark, it is common to pay a small amount for even tap water, which was definitely a bit of a culture shock coming from Canada, though a lot places will give it to you for free when ordering food, or at least hide the cost in the bill.
@ImpossibleScott I wish that someone would get customized personal shared virtual spaces right (w/o spatial locality extortion that we see in crypto based environments which does not make any sense for virtual worlds). Ernest Cline did such a great job depicting what it will eventually become.
made a super simple first-person example for everyone to remix. It's nothing much, but it's so satisfying to create stuff like this in minutes from a single app๐
#madeWithUnbound#IndieGameDev
@matiasgoldberg Yeah, I wonder if this method is exploiting the low flicker fusion threshold in the periphery outside of the fovea so any pixel deltas between frames will be easier to notice.
@matiasgoldberg@benhouston3d Until VR and AR resolve all disparities especially with regards to eye view parallax and also accomodation, it will continue to physically strain users during prolonged usage periods.
@matiasgoldberg@benhouston3d There was a very recent publication which showed VR can alter our visual consistency in relation to VR motion sickness. Any heavy continuous disparity between vestibulation, proprioception, and visual acuity is going to impact our perception IRL.
@SebAaltonen It is really unfortunate that people are being oppressive with machine learning tech. It should never be forced onto a problem and should only be used when it actually has an improvement over traditional methods, which is generally not the case.
@SebAaltonen Though the comment about screen space limitations is not accurate. That is exactly where generative ML becomes interesting when there is not enough data to extrapolate. Stable diffusion may be image space but it definitely can have an understanding of occluded elements.
I just released my first hugging space online demo, #ViewDiffusion which is a small piece of a larger local demo. This online demo is using stable diffusion outpainting out of the box to expand your field of view and generate an alternate local reality:
https://t.co/y9INlYkOTp
I've made a lot more improvements/changes to the controls, noise scheduling, and image streaming in my demo. The quality has significantly been improved. All that is required now is more GPU resources to approach real-time framerates.
@idoccor It is still only around a frame per second at best at a low number of inference steps. It would be interesting to see how smoothly the demo runs with an A100. I still have some room for potential optimization on my side. @huggingface
I've made a lot more improvements/changes to the controls, noise scheduling, and image streaming in my demo. The quality has significantly been improved. All that is required now is more GPU resources to approach real-time framerates.
I ported the latent walk control from my local demo to the online one, and it is live. At startup, it simply generates two sets of random latents as suggested by Karpathy in their gist. The "Latent Walk" value can be used to explicitly lerp between the two latents.