TDA3MV :IPIPE Gamma process

TDA3 is being used in many IR sensing applications. Image pipe in this case remains the same as the one for visible, but the processing for CFA interpolation, color processing etc need to be bypassed or tweaked.

In the standard image pipe for visual processing, 12b needs to be tone mapped down to 8b because most display panels can handle only 8b data. If you are using ISP output only for machine vision, you may tap out at IPIPEIF stage and bypass IPIPE. You will most probably not need any color/noise/sharpening etc. so you will not lose anything by bypassing IPIPE.

Thanks for the reply

I will try to explain again with pictures.

As shown in the picture above, data is input in 12 bits. After that, it is output as 10 bits through the gamma table.

Even if it does not pass through the gamma table, it is shifted by 2 bits.
The picture below is when the gamma table is bypassed.
When you look at the monitor, you see a contour line.

Are you using the standard 12b RAW Bayer dataflow? That will not give best image quality for 12b grayscale image. You will need to tune CFA coefficients and other ISP parameters (like disable color processing etc.)

You may also need to tune gamma to get lowest level of contouring.

Thank you for your answering.

But I am not good at TI-SOC such as you.

So I'am glad if you explain details.

Question0.

I don't understand why i can't get best image in 12b grayscale image.

I set the color processing to the disable mode.(first and secondary RGB2RGB)

Could you let me know how to set CFA coefficients with details?

(I dont have any clues about it)

Question1. 

The input format of IFIFE moudle is 12bit grayscale image. 

Below, I got answer in this forum for gamma module.

I have a question about 12-bit input format.

We get 12-bit data of the camera. But I think that the gamma module use only the msb 9bits in 12-bit data.

What is the meaning of the 0x0005(the lsb 3bits) in the value 0x0B15?

Could you let me know how to make 12 bit format

------------------------------------------------------------------------------------------------------------------

For example, if you have a 12-bit input data with a value of 2837, which is 0xB15 in hex value, you'd first down-shift the data by 3 to get the index to the gamma look-up-table, which is 0x162 in this case, or 354 in decimal.The lsb 3bits of the data, which is 5, is upshifted by 3 to yield 40.

Now assume your  gamma[354] entry has a offset value of 719 and a slope value of 2.

The final output from the gamma table operation will then be:

clip(719+((2*40)>>6)) = 720.

-------------------------------------------------------------------------------------------------------------------

Question2.

So, i make up 12 bit data that the msb 9 bit data is camera data and the lsb 3 bit is 0.

In this case, there is contour line in image, too.

I think that this case is caused by the lost of lsb 3 bit.

Am i wrong? If then, Could you let me know the reason? or How can I solve this problem?

Question3. 

When I use the gamma module, the contrast of the image get lower compared to not using gamma module.

i think that this problem is caused by the clip, 12bit to 10bit. 

Is this right? Then, how to i can solve?

Actually, I need to used edge enhancer in IPIPE module.

So I have to use gamma module in IPIPE module.

If you have other solution, then let me know it, please.

Thanks,

Can you please break different questions in different threads?

For CFA tuning, please refer to the document

Question 1:

For example, if you have a 12-bit input data with a value of 2837, which is 0xB15 in hex value, you'd first down-shift the data by 3 to get the index to the gamma look-up-table, which is 0x162 in this case, or 354 in decimal.The lsb 3bits of the data, which is 5, is upshifted by 3 to yield 40.

Now assume your  gamma[354] entry has a offset value of 719 and a slope value of 2.

The final output from the gamma table operation will then be:

clip(719+((2*40)>>6)) = 720.

Your understanding is correct.

When I use the gamma module, the contrast of the image get lower compared to not using gamma module.

i think that this problem is caused by the clip, 12bit to 10bit. 

Is this right? Then, how to i can solve?

Can you please share the gamma table you are using?

Please also share the 12b image you are using.

The 12-bit BGGR format is used.

When I checked the TRM, I saw the CFA module output in RGB-4:4:4 format. Could this be the cause?

Currently, it is being tested with YUV422 input as shown in the figure below.
Testing in the order of IPIPEIF->RSZ.

However, if you look at the register description field, VPORT does not have YUV input. It turns out that YUV is possible only in SDRAM.

Can you tell me how to write YUV input through SDRAM?

Hello Seok,

To be able to support you better I will need the answer to previous questions. I had requested gamma LUT and the image you are using with which you are g etting contours. Please respond to that first.

Questions on CFA or resizer should be in a new thread. It becomes chaotic to mix several issues in one thread.

Regards,

Mayank