ISL and Collaborators
2 HN recommenders · recommended in 7 months, 2021–2025
8:34 · 6.3M views · May 10, 2021
A video presenting an Intel research paper on enhancing the photorealism of synthetic images. A convolutional network uses rendering intermediates, shown on GTA V footage that comes out looking like real street scenes.
Used as a go-to example of neural networks enhancing game rendering, referenced in later AI-graphics discussions.
10 comments from Hacker News
DLSS & ideas like it outside an idea are really interesting: [LINK] Ex. run the base game at an insanely crummy/low resolution. On low(er) it’s possible to run a model to kinda “upscale” in each frame in a performant way. Or even this CNN initiative from Intel where they took Grand Theft Auto & “painted” it with dashcam footage from rural Germany: this video The results are pretty interesting— all “brown” southern CA mountains were green like in Germany
The video linked of applying a neural net to GTAV for photorealism is really impressive if you haven't seen it. this video
Video is really a series of frames, the framerate for film/human can get away with 24 frames/second-- so maybe ~40ms/image for real-time at least? What's cool about the era in which we live is if you look at high-performance graphics for games or simulations, for instance, it may in fact be faster to a the model to "enhance" a low-resolution frame rather than trying to render it fully on the machine. ex. AMD's FSR vs NVIDIA DLSS - AMD FSR (Fidelity FX Super Resolution): [LINK] - NVIDIA DLSS (Deep Learning Super Sampling): j [LINK] AMD's approach renders the game at a crummy, low-detail resolution then each frame uses "upscales" Both FSR and DLSS aim to improve frames-per-second in games by rendering them below your monitor’s native resolution, then upscaling them to make up the difference in sharpness. Currently, FSR uses spatial upscaling, meaning it only applies its upscaling algorithm to one frame at a time. Temporal upscalers, like DLSS, can compare multiple frames at once, to reconstruct a more finely-detailed image that both more closely resembles native res and can better handle motion. DLSS specifically uses the machine learning capabilities of GeForce RTX graphics cards to process all that data in (more or less) real time. Video is really a series of frames, the framerate for film/human could get away with 24 frames/second-- ~40ms/image for real-time. What's cool about the era in which we live is if you look at high-performance graphics for games or simulations, it m
> AI will be doing graphics programming? I doubt that. I think the bigger concern (if you could call it that) is that certain AI techniques may be able to replace entire stacks once they get advanced/stable enough. Something along the lines of this perhaps, where you render simplified geometry and let the AI fill in the blanks: this video Even without experimental technologies, you can get a glimpse of how shipped tools like ray reconstruction can morph the field in the future, forever entrenching themselves into graphics programming one way or another. As far as AI writing graphics code? No way, at least not in the next couple of decades. Button snippets are a far cry from rendering virtualized geometry 60 times a second.
Researchers trained a deep learning network on photos to make GTA5 more realistic-looking: this video I was surprised that most of it is is just tone mapping. It looks too green-ish to me, needing some white balancing, but realistic. I think Video games are intentionally over-saturated and contrasty to make them look better, just as paintings are[1]. I took many screenshots in Horizon Zero Dawn that would be look bland if it was realistic like the modfified GTA5 footage above[2]. You also don't want animation movies[3] with realistic tone mapping. But for a horror game you would do exactly that, it's about immersion (or you would never be frightened), not beauty. [1] [LINK] [2] [LINK] [3] [LINK]