Best way to fine tune LUT

[s1mptom]

Hello,
I’m seeking advice on the best way (or application) to quickly and conveniently prepare a LUT file. I seem to be having some difficulties fine-tuning the picture using the built-in configurator. I haven’t been able to create a stable version where colors in both bright and dark scenes are displayed as close to accurate as possible. I’m using hyperion-webos as the image grabber on an LG G3 TV, with an Nvidia Shield as the source for Dolby Vision content. Adjusting the picture for bright scenes was relatively straightforward, but for some reason, dark scenes tend to shift towards blue-green tones. This might be artifacts from the grabber, or perhaps something else.
I would like to try preparing a LUT file with a bit more control over the settings, but other tools seem a bit too “professional” for my needs.

[awawa-dev]

Hi
This is no longer Dolby Vision but some signal converted in pain to RGB for internal webos needs. You would have to ask LG engineers why it came out so poorly or what webos capturing (powered by reverse engineering) is doing wrong. Maybe someday I will implement finally automatic calibration thanks to AI and the situation will improve a little but you should not expect miracles because we do not know what we are dealing with.

[awawa-dev]

This template should be extended for image format type on both sides:

HyperHDR/include/flatbuffers/parser/hyperhdr_request.fbs

Line 9 in c604658

table RawImage {

Then we can go to implement sending (I dont know webos capturing so cant help here) & decoding (NV12 is already supported but the flatbuffer server doesnt use it yet): https://github.com/awawa-dev/HyperHDR/blob/master/sources/utils/FrameDecoder.cpp
BTW small LUT table 50MB (instead of full one: 150MB) wont work anymore because it is missing the table we need for YUV & tone mapping decoding.

[s1mptom]

Oh, you’ve given me an interesting task/idea! I’ll dive deeper into how hyperion-webos and flatbuffers work and try to implement the sending of NV12. I’ll start by setting up a test server for debugging. As for the small LUTs—I don’t have any of those, only the large ones through the built-in calibration. I’ll be back when I have more questions :D

[s1mptom]

It looks like I've managed to send the original PIXFMT_YUV420_SEMI_PLANAR(?) image to my debug server. The image is, of course, dark even in SDR—probably because it's missing the conversion coefficients, but it's a start! To set up the stream reception on the HyperHDR side, do I just need to update the flatbuffers table, or is there anything else required?

My new table for sending (and receiving) looks like this:

namespace hyperionnet;

// A priority value of -1 clears all priorities
table Register {
  origin:string (required);
  priority:int;
}

table RawImage {
  data:[ubyte];
  width:int = -1;
  height:int = -1;
}

table NV12Image {
  yData:[ubyte];
  uvData:[ubyte];
  width:int = -1;
  height:int = -1;
  stride_y:int = -1;
  stride_uv:int = -1;
}

union ImageType {RawImage, NV12Image}

table Image {
  data:ImageType (required);
  duration:int = -1;
}

table Clear {
  priority:int;
}

table Color {
  data:int = -1;
  duration:int = -1;
}

union Command {Color, Image, Clear, Register}

table Request {
  command:Command (required);
}

root_type Request;

Is using the NV12Image type correct? I noticed that the FrameDecoder uses just NV12—do I need to update this in the table and rebuild? Additionally, since I implemented this sending "on the side," parallel to the main one (going through the converter), I only implemented the frame sending and nothing else. Will this be sufficient, or do I need to implement any additional functions related to message processing on the HyperHDR side?

So, I want to understand what the next steps are that I need to take. Or if HyperHDR receives an NV12 image, will it handle everything else automatically?

[awawa-dev]

Great! Could you send me on email (the address is in my profile) size of the sample nv12 frame in bytes, width, height, stride_y, stride_uv? And if possible, an example of this nv12 frame in raw/binary form or encoded in flatbuffer format?

[s1mptom]

I’ve sent a bin file containing the y and uv data, hopefully to the correct address. Additionally, I’ve included the server log output with the message size and parameters (I hope it contains all the necessary information). It seems like the player “went to sleep” and reduced the brightness, so I’ll resend the captured frame just in case.