This script helps identify correct fix for a tint difference between two sources. It first attempts to apply the following common errors:
- 2.2 / 2.5 QuickTime "gamma bug"
- TV to PC levels
- PC to TV levels
- 0-235 / 0-240 levels
- 16-255 levels
- BT.601 to BT.709 color matrix
- BT.709 to BT.601 color matrix
- 10-bit to 8-bit truncation
If fixing these does not result in low enough mean differences, it falls back to attempting to identify the tint with matchcolors with the following parameters:
- gamma
- offset and gain
- levels
Keep in mind that the results from matchcolors are to be taken with a grain of salt.
The script tends to predict gamma most commonly and struggles the most with level adjustments.
If results are unsatisfactory, it can help to manually run matchcolors with different params or just accept that the grade isn't fixable by a script this simple.
When passing YUV video, autocrop is applied with standard ranges. RGB images must be cropped properly before passing to the script. Both black bars and texts need to be cropped.
To run the script on two files that are identical apart from their tint:
python -m gettint path/to/tinted/file path/to/reference/file [--frame FRAME_NUMBER --force-matchcolors]The --frame option is only relevant for videos.
If you know you aren't dealing with one of the common errors listed above and would like to skip those, set --force-matchcolors (or -m).
This is a super basic script to aid with detinting sources that have had simple color corrections or level adjustments performed on them.
For this to work properly, black bars must be cropped, and clip contents must match exactly. A low-pass filter is applied by default to help with noise-related fluctuations.
To e.g. test for gain and offset in RGB in my_script.py:
from gettint.matchcolors import matchcolors
matchcolors(image, reference, params=["gain", "offset"], format=vs.RGB24)Then run the script and look at the output as well as the graph found in fig.pdf:
python my_script.py- IRC Channel:
#OpusGangonirc.libera.chat