Closed j2l closed 1 year ago
I corrected the perspective and it keeps on wrongly interpreting the same colors (first 4 and first 3 greys).
cropping more and saving as JPG seems to fix it for colors: interpreted as:
But since color cards are usually with borders, a way to see/select areas is still needed (area to average noise, especially in low light). Too bad I can't code in python, I'd love to improve it with an array of rectangle you could move manually.
BTW, it should be mandatory to upload a photo taken at the same moment to test the correction instead of using your pic since your camera and light condition is different from mine :)
Thank you for the feedback. I can probably make a mode for color-cards that accounts for borders, or a way to select the points. It might be easier for me to make that in Javascript, to pre-process the image, rather than in Python though.
BTW, it should be mandatory to upload a photo taken at the same moment to test the correction instead of using your pic since your camera and light condition is different from mine :)
Yeah, definitely don't use my image if using your own color cards - that wouldn't make any sense.
I modified it a bit, adding a border-value option now. I have it pre-set to 25%, which should help I think in your case.
I also ran script with my own color card, and it turned out pretty okay: https://github.com/steveseguin/color-grading/blob/master/spyder_24_color_card.ipynb
If there's still an issue with noise, I suppose I can add a denoiser , but a blur filter might also work.
Great! Thank you @steveseguin !! Funny, looks like we have a mirror of each color card.
Your camera looks far better than mine though, histogram doesn't have more than the 3 RGB values, like I do. Since I have this "noise" value on the left side, and it changed after i cropped, I guess I would need blur. I could add it when fixing the perspective.
I'll test it tomorrow, thank you again!
I bought my card directly from Datacolor; wasn't cheap of course, but since I bought it as a replacement card (without case), it was reasonable.
Histogram is now far better. It works fine, so I'm closing this ticket.
Thank you again for the fix and the overall quality of this project. It encourages me to get a decent camera :smile:
Fantastic project to learn calibration @steveseguin, thank you!
I have a color reference card with known RGB values so I need exactly this project to make a correction LUT in OBS. I shot a full size ref card with some slight perspective and it also have numbers. Maybe because of this, it doesn't see the first colors and I get more RGB values in histogram and computed LUT is incorrect .
Is there a way, in Step 4, to see/tweak which point/area it "sees" (uses)? Here are my files enclosed to get an idea. expected:
ref card