TDA4VM: TDA4 Decompading issue on isp tuning

Hi renhui,

I am not sure what you mean be "same" problem.
You may check out the python script for setting up default DCC profile including decompanding here -- https://git.ti.com/cgit/processor-sdk/imaging/tree/tools/default_DCC_profile_gen?h=main

BTW, latest DCC tuning tool is V3.1 now.

Hi Gang,

Thanks for your quick relpy. I'm sorry I didn't describe my problem clearly.

My problem is that I used the decompanding curve provided by sensor fae to decompress pwl12 to 24bit, but the brightness of the  image is very dark, and the adjustment of exposure gain is still dark. However, the brightness of the decompamding curve by default using dcc tool is normal. May I ask what was done in the process of pwl12 decompression to 24bit and compression to 16bit, and how to troubleshoot this problem, thank you

Best regards 

Renhui

Hi renhui,

You may use the python script to set up decompanding for your sensor and choose a proper gamma value for compressing to 16-bit in VISS.
Please follow the instructions in the sensor property file -- e.g., https://git.ti.com/cgit/processor-sdk/imaging/tree/tools/default_DCC_profile_gen/configs/imx390_properties.txt?h=main
The gamma value is typically around 0.48 ~ 0.5 for 24-bit sensors.

Because your sensor is 24-bit, you will need enough exposure in the sensor for dark scenes to get bright output image at night scenes.
In my experience with 24-bit Sony ADAS sensors, we use fixed 10ms exposure time and up to 24dB sensor gains for dark scenes.

BTW, which sensor are you working on?

7752.XML.zipHi Gang，

The sensor is ox08b. According to your suggestion to adjust the parameters, found that the effect is still very poor. When I run to the maximum exposure and maximum gain, it is still all black in the room, except for the lamp tube. I upload the xml file, can you help me check whether the parameters of wdr and H3A modules are reasonable? thank you!

Bese Regards 

Renhui

I modified the compression curve of the sensor to be linear, from 24bit to 12bit, and the curve of his decompanding was also linear, from 12bit to 24bit.

Hi Gang，

May I ask if you have a mature IQ para ？  Could you provide these   for my reference? Thank you！

renhui said:

May I ask if you have a mature IQ para

We have not worked with any X8B camera at TI, although some of our customers have.

Hi renhui,

renhui said:

According to your suggestion to adjust the parameters, found that the effect is still very poor. When I run to the maximum exposure and maximum gain, it is still all black in the room, except for the lamp tube.

Are these xml files coming out of the python script with a pre-compress gamma value around 0.5?

What are your maximum exposure time and gain values?

Please also share with me a sample raw image along with your sensor companding knee points captured with the max exposure time and gain.

Hi Gang，

I'm sorry that I didn't reply you in time because I need some urgent things these two days. I used a new set of IQ parameters and set gamma to 0.5 according to your suggestion. This problem has been improved. I suspect that the previous generate_dcc.sh did not compile all parameters into so, resulting in some parameters using default. Could you please help check and find out root casue？ The attached file is tow group of xml para and the question raw and sensor pwl12 curve. Looking forward to your reply, thank you!wdr_decompanding_question_data.zip

Best Regards

Renhui