The issue is getting an AI to do this on more complicated (and arbitrary) sets of data. If I gave you a minimalist company logo at only 16x16 or 32x32, chances are you could create an accurate vector drawing of that logo that looks perfect scaled up to 5000x5000, because logos are typically very simple shapes put together in very simple ways so despite having a TINY explicit information set, you can tell from the countours and designs with extreme accuracy what the image would look like upscaled to an arbitrary resolution. The reason we use AI in these areas is because they can become very good at sampling implicit information in arbitrary examples. It'll never be a perfect match to what the media would be if it were native obviously but the assertion that "the information isn't in the file so there isn't a point" is only going by explicitly declared information, much like only drawing a picture on-screen based on the pixels themselves. In both cases your dealing with using an AI to estimate additional data not explicitly stored in the original media, rather using implicit information (such as "sub-pixel" information in upscaling, contours in the scene, etc.) in order to get a good idea of what it WOULD be. If you have a GPU and up to date Kdenlive, render with NVENC H265 under "GPU" format.Ī year old but this is basically the same as AI upscaling just in a different mode. If they look the same or very close, you're done. ![]() Once you're done, you should compare with your SDR device. For the overrides or inputs, I keep as is but I change color range to mpeg partial. You may experiment with this, but I go with REC 2020 as the output on the Color Matrix, I use the "colour correct" LUT, then I set all Color Space output to REC 2020(I go with 10 bit) and PC full range with dithering. Then in Kdenlive, I click timeline, go to effects and apply (in this order), Color Matrix, Apply LUT, Color Space. I got mine by emailing the 4K HDR Anime Channel. My solution: Use your phone as an SDR display and as reference. The weird colors may highlight artifacts. Compare the video with an SDR display and HDR display. The HDR conversion looks a bit off while on SDR mode compared to real SDR, but it looks good on HDR mode compared to the oversaturated abomination Windows 10 was displaying beforehand. This makes it so whether on SDR or HDR it looks the same-ish. ![]() I've been experimenting with LUTs and color matrix and spaces, but I really don't know what I'm doing.ĮDIT: While not true HDR, I was able to convert SDR colors to HDR colors. Is there a method that can simulate HDR from SDR? At least increase darkness and brightness?
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |