@bingyikang@huggingface@_akhaliq@pcuenq@NielsRogge Wow. Iโm curious how the demo videos were created? as in terms of visualising the reconstruction as it roams around the building, or the first video clipโamazing!
Google presents Diffusion Models Are Real-Time Game Engines
discuss: https://t.co/IBV70a990g
We present GameNGen, the first game engine powered entirely by a neural model that enables real-time interaction with a complex environment over long trajectories at high quality. GameNGen can interactively simulate the classic game DOOM at over 20 frames per second on a single TPU. Next frame prediction achieves a PSNR of 29.4, comparable to lossy JPEG compression. Human raters are only slightly better than random chance at distinguishing short clips of the game from clips of the simulation. GameNGen is trained in two phases: (1) an RL-agent learns to play the game and the training sessions are recorded, and (2) a diffusion model is trained to produce the next frame, conditioned on the sequence of past frames and actions. Conditioning augmentations enable stable auto-regressive generation over long trajectories.
deepmimic style transfer in unity for physically-based animations.. something so uncanny about observing physically 'accurate' humanoids.. slowly developing some sort of weird empathy for these machines๐ธ
@carlosedubarret@poomnattawat1 Ah yes! I was looking in that folder last night. Their https://t.co/3ex0MdyfWY file is a direct copy of the main fn found in 4dhuman. Iโll try this and see
@carlosedubarret@poomnattawat1 How do you install SLAHMR with 4dHuman? Iโm able to install both separately. As I see in their latest update they support 4dh, but Iโm not sure how to put both together. Or am I misunderstanding how SLAHMR works? Thanks for the sick work!
@dvsch I don't think he is ignoring/disregarding current issues that plague AI with this post. He is just thinking on a grander timescale - nothing wrong with that imho. I'm sure he has talked about issues before such as 'bias' in his countless talks, and has a good understanding of it
There are a bunch of papers following this one. NeRFs soon will be moving and all sorts of interactive stuff.
Like moving a car around a street kind of weirdness.
Finally ML has managed to match the SHRDLU system that Winograd wrote in 1968. It took over 50 years but now it works in the real world and is learned from data.