![]() You need to be careful how you train it, or you might bias the algorithm. Thats really the hard part with training (outside the computational requirements). Low-quality content does not qualify for that particular type, since even if it was downscaled once, those attributes were destroyed by over-compression, noise, or whatever makes it "low quality". If you do such a training, the filter doesn't learn to "upscale", it learns to un-do the downscale - which in theory sounds similar, but in practice can end up quite different. If it was truely trained on taking high-res originals and downscales of those, then that is entirely expected. It should have happened a long time ago because grading content to 100 nits doesn't make much sense for almost all current displays. HDR still has a long ways to go to live up to the promises made by Dolby. But what can you do.but buy a brighter display or a laser projector. I like the picture produced by HDR -> SDR, but I can't help but notice the lack of visible steps when you lower the target nits and brightness excessively. The only downside will be the number of frustrated people who want to understand how to use it all. ![]() This data also makes it possible to cherry-pick scenes as demo material. I have seen mistakes made in reviews where a reviewer has erroneously assumed one display is not any brighter than the other while comparing content that doesn't reach past the peak brightness of either display. There are professional TV and Blu-ray reviews out there who would love to have this information. Even though most projector owners can't see past 100 nits, they are now analyzing real-time graphs of brightness histograms and preparing weighted averages of HDR content based on the measurement files produced by LAV Filters and madVR. The level of nerdiness with the tone mapping is becoming impressive. Madshi will have a new release out soon with some big improvements to HDR tone-mapping. ![]() I have seen some content look like an oil painting that kind of made me drunk while watching it. With that said, I don't think NGU Sharp was trained well with poor material. All of those eyelashes add up when the image is large. Lanczos creates some artifacts, but it also smears a lot of valid detail. If, like me, you are frequently comparing Kodi VideoPlayer with Lanczos3 to NGU Sharp very high, they aren't the same image. Personally, I think all images should eventually be scaled by neural networks when Tensor Core technology is scaled down. NGU Anti-Alias is also a very similar upscaler to NNEDI3 and most seem to prefer it. NGU very high pushes the GPU so hard that hundreds of thousands of repeated calculations must be taking place. If you look at screenshots, you can see that NGU picks up details like eyelashes that are missed by Lanczos and super-xbr. Maybe he has an evil lair with a bunch of super computers/GPUs to do this type of training? madshi posted at some point that the training was done by comparing downscaled images to the original. It was mentioned a while ago that NGU is based on neural networks. Of course no mere mortal has that sort of resources to train a network. All those fancy NVIDIA demos you see for image processing, those are trained on super computers, large clusters with hundreds if not thousands of NVIDIA GPUs to train the network. ![]() You need absolutely massive compute power to train a Neural Network. Neural Networks can be really powerful, and who knows if some day one might return to madVR, but the real challenge with Neural Networks is training them. But it was removed because it was ultimately decided that the added complexity in madVR was not worth it (since it was the only component to use OpenCL IIRC), and other algorithms could flat out replace it at higher speeds and quality. We had NNEDI3 in madVR for a while, a Neural Network based scaler. Not to mention even newer stuff with specialized hardware just for that task, ie. Wut? Most Neural Networks rely on GPUs to run them at any speed, because GPUs are actually pretty good at that, because it can be paralized really well.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |