Tool/software:
Hello,
SDK used: j784s4 version: 09_02_00_05
Camera Used: Onsemi 341 MARS and 341 A Sample (Stoneridge designed)
We recently had a call with TI (Gang) and received some insights on general tuning procedure for AWB, and the conclusions of the meeting were:
1. H3A will subtract WB gains whichever was applied before it and collects the data.
2. AWB should be tuned on the data that does not have any WB gains.
We did the following experiments:
1. Collected Raw images with pre-gain enabled from 341 A Sample camera to tune AWB and CCM. Observed in the output for 341 A sample camera all the white color is in green as seen in the image below:
2. Collected Raw images with pre-gain enabled from 341 MARS camera to tune AWB and CCM. Used this tuning output to run 341 A sample camera and observed the results are as expected.
3. Collected images with no Pre-Gain applied and tuned AWB and CCM. used this tuning for 341 A sample camera (Stoneridge design) and observed the output is as expected.
Furthermore, we confirmed no gains are applied for the two WB blocks mentioned as per RAWFE pipeline in the below image, highlighted in the red boxes:
1. For Wb1 we checked the structure: Rfe_Wb1Config within rawfe_cfg.h is not used anywhere and ascertained that the wb1 is not being applied.
2. For Wb2 we printed out the digital gain value per tiae_exp_t structure used in tivxAewbProcess function, and verified the digital gain being applied is 256 (which as per the definition is 1.0x)
3. through the above mentioned two steps we confirmed Wb1 and Wb2 are not applied in the pipeline. Please confirm if this is enough to ascertain regarding these two blocks or if there's is any other place these can be configured.
As per discussion with gang and based on our conclusions that WB1 and WB2 are not applied, and the AWB tuning performed with the RAW images that didn't include any WB gain (Experiment 3), produced the results as expected, can we assume H3A does WB subtraction on the data if at all it is applied before it reaches H3A. However, as per the comment mentioned in the definition for Rfe_wb1Config structure (line 315 in rawfe_cfg.h, highlighted in the image below) H3A received data is not WB corrected. Does this mean H3A generates stats for the data that has gains applied or is this gain subtraction happening elsewhere. If the gain is removed from the data before H3A, kindly let us know where it is performed.
In addition to this, although experiment 3 mentioned above yielded expected results. The Awb tuning performed in DCC for these images did not result in linear graph tuning output, as mentioned in this TI ticket: TDA4VH-Q1: AWB Tuning for AR0341 sensor - Processors forum - Processors - TI E2E support forums. What could be the reason for the dip observed in the tuning output, and what is typically the impact of this dip, we are confused because, despite the fact that AWB tuning is not proper in DCC we still see proper output.
Please provide answers to all the questions asked above and let me know if you need anything more for you to help assess the situation better.
