• State
  • Locked Locked
  • Replies 15 replies
  • Subscribers 101 subscribers
  • Views 818 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AM62A7: AWB calibration

peng qiang
Intellectual 310 points
Part Number: AM62A7

Hi    TI experts 

Is there a relationship between the calculation of gain in the AWB algorithm and the color temperature displayed on DCC?

I found that when the color temperature on DCC changes from 3000 to 8000, the gain value remains basically unchanged.

  • 0
    TI__Guru 58940 points

    Hi peng,

    peng qiang said:

    Is there a relationship between the calculation of gain in the AWB algorithm and the color temperature displayed on DCC?

    Yes, there is not a direct impact, but gain and color temperature should have some correspondence.

    peng qiang said:
    I found that when the color temperature on DCC changes from 3000 to 8000, the gain value remains basically unchanged

    AWB gets the color temperature information from your calibration input for each reference raw file.
    You will have to input it correctly for each of you raw image files.

    if you manually change it from 3000 to 8000, gain may not chance in certain conditions, but it still may mess up AWB in calculating the color temperature.

  • 0 in reply to Gang Hua
    Intellectual 310 points

    Hi  Gang

    I am not manually modifying the color temperature.When I debug H3A online, the screen changes from light to dark, and the color temperature in DCC changes from 3000 to 8000. At this time, the scene remains unchanged and the gain remains unchanged.

    There is currently an error in DCC color temperature calculation after calibrating AWB. For example, when the color temperature was originally 8000k, DCC only displayed 5000K.

    I would like to ask if the calibration of AWB is accurate based on the ineffective CFA module.

  • 0 in reply to peng qiang
    TI__Guru 58940 points
    peng qiang said:
    When I debug H3A online, the screen changes from light to dark, and the color temperature in DCC changes from 3000 to 8000. At this time, the scene remains unchanged and the gain remains unchanged.


    How light and dark are the scenes?

    You may check if H3A output is too low (e.g., all 0s) in the dark case.

    peng qiang said:

    There is currently an error in DCC color temperature calculation after calibrating AWB. For example, when the color temperature was originally 8000k, DCC only displayed 5000K.

    This is not expected if AWB calibration is proper.
    Can you share your AWB calibration output plot, xml file, and the 8000K raw image (and knee points)?

    What type of sensor is this?

    peng qiang said:
    I would like to ask if the calibration of AWB is accurate based on the ineffective CFA module.

    I am not sure if I understand this question.
    What is "ineffective CFA module"?

    AWB takes input from H3A only.

  • 0 in reply to Gang Hua
    Intellectual 310 points
    Gang Hua said:
    Can you share your AWB calibration output plot, xml file, and the 8000K raw image (and knee points)?

    case.zip

    Gang Hua said:
    AWB takes input from H3A only.

    Before adjusting AWB, as long as H3A is determined and the brightness of the screen is appropriate, it is sufficient for AWB.

  • 0 in reply to peng qiang
    TI__Guru 58940 points

    Thanks for sharing!

    AWB calibration looks reasonable (references are from 7419K to 2399K.

    What is the black level of "8000K.raw" (after linearization)?

    Is it actually under 8000K lighting?
    If yes, it is beyond your AWB calibration range.

  • 0 in reply to Gang Hua
    Intellectual 310 points

    Hi Gang

    Gang Hua said:
    What is the black level of "8000K.raw"

    64

    Gang Hua said:
    Is it actually under 8000K lighting?
    If yes, it is beyond your AWB calibration range.

    The illuminometer displays close to 8000.

    The file I gave you was reproduced by me. The actual effective AWB file has a color temperature of 8K during calibration.

  • 0 in reply to peng qiang
    TI__Guru 58940 points

    This 8000K raw image is not consistent with your calibration xml file (not on the calibration curve).
    AWB may have trouble guessing the color temperature in this case.

    Is that 8000K light one of your calibration lights?

  • 0 in reply to Gang Hua
    Intellectual 310 points

    This is all the Raw images I use.

    The current problem is not that color temperature calculation errors only occur at 8000K. Even at 7000K, the calculated color temperature is over 5000.

    I gave an example of 8000K because there is a greater difference in color temperature at 8000K

  • 0 in reply to peng qiang
    TI__Guru 58940 points

    Hi peng,

    Is that 8000K light (8000K.raw) one of your calibration lights?
    Do you have this trouble with all of some of your calibration lights?

    I would need to understand your situation better.
    So, please clarify.

    In general, for all your calibration lights, AWB should work properly and give expected color temperature (as you specified in tuning tool) for all your calibration lights.
    If you test with other lights out of the calibration range, AWB may have trouble.

    It is not clear to me if your tests are for calibration lights or others.

    The 8000K.raw image is clearly out of the calibration range in your xml file given your shared "asamplesensor_kneepoints.txt" (all in your 10.18 case.zip).
    If that 8000K raw image is in your calibration lights, then your "asamplesensor_kneepoints.txt" could be wrong.

  • 0 in reply to peng qiang
    TI__Expert 4780 points

    Hi, Peng

    Would you please share all the calibration image used. 

    Thanks!

  • 0 in reply to peng qiang
    TI__Expert 4780 points

    Hi  Peng

    Thanks so much for your sharing!

    I think the 8000K image could not be 8000K. While I put all the images into DCC tool, the 8000K image is not consistent with other images:

    So there must be something wrong with your 8000K image, either the light source or the illuminometer .

    How was the 8000K produced? combining other light source used in producing other normal images? or involving a new light source?

  • 0 in reply to Adam Hua
    Intellectual 310 points

    Hi Adam

    This 8000K Raw image is multiplied by gain (calculated gain when DCC displays 5000).

    The production environment of 8000K is the same as that of the calibration image, both in the lightbox.

  • 0 in reply to peng qiang
    TI__Expert 4780 points

    Hi Peng

    do you mean that the image 8000K has been applied with awb gain? If so, please  provide an image under 8000k without awb gain.

  • 0 in reply to Adam Hua
    TI__Guru 58940 points

    Hi peng,

    peng qiang said:

    This 8000K Raw image is multiplied by gain (calculated gain when DCC displays 5000).

    This is why we have been confused.

    In general, AWB shall work properly for the calibration scene if all parameters are entered in tuning tool and SDK source code properly.
    For applying AWB gain in the sensor, you will need to set "sensor_pre_gain" to 1 for AWB.

    I also created one AWB calibration project below using your raw images, which you may check against your project.
    I assumed a gamma value of 50 and black level of 64 in the project (I am not sure if they are correct for you, but calibration output plots look good to me).

    https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/791/test_5F00_AWB_5F00_X3C.7z