TDA4VM: dcc_gen issue

Hi Cherry,

I'm trying to understand more on this request so as to assign this to the right expert?

Can you please expand on the request?

Regards

Karan

Hi Karan,

Thanks for your help.

For PROCESSOR-SDK-RTOS-J721E_7.0.1, the psdkra/imaging/tools/dcc_tools/dcc_gen_linux file in the sdk directory is an x86-linux architecture. Could you provide an executable for the a72-qnx architecture? Or the code for dcc_gen_linux?

Thanks and regards,

Cherry

Hi Cherry,

This tool is used to convert DCC xml files into binary format and this is offline tool available for Linux. It is not run/supported on target. Why exactly do you want to run on target or on QNX?

QNX or Linux running on target does not really require to run this tool. They will just parse the binary file and use ISP parameters.

Regards,

Brijesh 

Hello Brijesh,

Thanks for your help.

Due to errors in the actual parameters of the same lens, this binary file needs to be dynamically generated on the board to address the differentiation issue. Could you help provide the conversion code? 

Thanks and regards,

Cherry

Hi Cherry,

Could you help me understand what exact parameters needs to be changed dynamically? On what condition is it going to be changed? 

Regards,

Brijesh

Hi Brijesh,

Brijesh Jadav said:

Could you help me understand what exact parameters needs to be changed dynamically?

In-lens reference: K1, k2, k3 etc.

Brijesh Jadav said:

On what condition is it going to be changed? 

It will be changed when the product is off-line from the factory or when the lens is changed.

Thanks and regards,

Cherry

Hi Cherry,

What is in-lens reference? Where is it used in ISP?

When lens is changed, it is recommended to change DCC bin file.. 

Regards,

Brijesh

Hi Brijesh,

The in-lens participation affects the results of LDC de-distortion, and since the actual parameters of the same lens have errors, different lenses need to be matched to different DCC bin files. To solve this problem, the customer need to dynamically replace the BIN file on the TDA4x platform, so they need the dcc_gen_linux executable under qnx or preferably convert the source code to integrate this functionality into the qnx version of process rtos sdk 7.0.1.

Thanks and regards,

Cherry

Hi,

May I know is there any update regarding the question above?

Thanks and regards,

Cherry

Hi Cherry,

For LDC, DCC file is optional, we can always call control command to update the LDC lookup table. Changing Lut for LDC is easier than porting dcc tools on the target.. 

Regards,

Brijesh

Hi Brijesh,

How to update the LDC lookup table with the command and how to update the LUT file?

Thanks and regards,

Cherry

Hi Cherry,

I think there isn't control command to update LUT, but i think it can be easily added, it would be easier to add this control command than to add DCC on target. 

Regards,

Brijesh

Hi Brijesh, 

Is there any way to update the LUT on qnx? An integrated approach  is preferred(for example, how to add control commands to update LUT).

Hi Cherry Zhou,

Do you mean the LDC LUT? Yes, once we have support for changing LUT parameters using Control command, it can be used even from QNX..

Regards,

Brijesh

Hi Brijesh,

So for now we are not able to support this functionality? 

Thanks and regards,

Cherry

Hi Cherry,

I will add this functionality and share it in this week.

Regards,

Brijesh

Hi Cherry,

Cherry Zhou said:

Due to errors in the actual parameters of the same lens, this binary file needs to be dynamically generated on the board to address the differentiation issue. Could you help provide the conversion code? 

Do you mean the the LUT will be dynamically updated when the usecase/graph is running? Because if this is not the case, we can always calculate LUT in the application, map the LUT/mesh buffer in the application space, copy the updated parameters and unmap the mesh image..  Does this look ok?

Even if the usecase is running, if the LUT is updated by mapping the mesh image, we might see one or two frames incorrect. But it would be ok once it is complete updated. In order to fix this, you could declare the mesh image as graph parameter and then can dequeue, update mesh image and enqueue it back..

Regards,

Brijesh