TDA4VM: Difference between single_cam app and multi_cam app on IMX390 DISCOVERY

Hi Zhou,

Xiaohan Zhou said:

We only enaled one camera for multi_cam app. And for images we took from these two apps, single_cam app always looks lighter.

Both apps should use the same VISS tuning as far as I can recall.
There could be some run-time difference in AE and AWB output if the camera is moved or the scene changed slightly.

Xiaohan Zhou said:

Can you tell me is this true, or what's really different if it's not.

I cannot tell from your pictures if that difference is persistent or not.
If you keep the camera/view exactly the same and repeat the experiment, do you see consistent image output for both apps?

To exclude the effects of AE and AWB, we force to set analog gain to 1024 and exposure time to 11000, color temperature to 4500, and WB gains to [820, 512, 512, 894].

Here is how I did:

And I have made sure those values are written to registers.

But their images are still different:

Another experiment is that we set skipAE and skip AWB to 1

And there are how they look like:

Hi Zhou,

Xiaohan Zhou said:

And I have made sure those values are written to registers.

What you did looks correct to me.
Are you able to confirm if the sensor raw images from single-cam and multi-cam are close enough?

Xiaohan Zhou said:

But their images are still different:

If the sensor raw images are almost the same, then VISS output should be as close in your case, unless one of the apps used different VISS tuning parameters from some DCC binary input.

Xiaohan Zhou said:

Another experiment is that we set skipAE and skip AWB to 1

This is mostly likely a difference in sensor output raw images, although I am not sure why exactly.

My friend Brandon Maciel  did a similar test on his EVM and found that the sensor raw image from multi-cam is brighter than single-cam.

Hi Gang
I think I found out the reason. 

There are two kinds of wdr config implemented for imx390 d3, 60fps for single_cam and the other one(30 fps) is for multi_cam. These 2 configs are quite different including AE settings.

There are 3907 bytes of register settings and some of them are modified which are even reserved by sony according to their datasheet.

so I couldn't understand this config all. Do you have a manual for this?

By the way, we want to know if we can control multi_cam or other object detection apps by DCC tool. Or should I propose a new question about this?

Best regards

Zhou

Xiaohan Zhou said:

There are two kinds of wdr config implemented for imx390 d3, 60fps for single_cam and the other one(30 fps) is for multi_cam. These 2 configs are quite different including AE settings.

The sensor settings should be mostly the same for 30fps and 60fps except for the frame rate if I understand correctly.

Xiaohan Zhou said:

so I couldn't understand this config all. Do you have a manual for this?

There are sensor details and you would have to get the details from Sony.

Xiaohan Zhou said:

By the way, we want to know if we can control multi_cam or other object detection apps by DCC tool. Or should I propose a new question about this?

No, that is not supported.
DCC only supports VISS and LDC.

The sensor exposure could be higher in multi-cam at 30fps than single-cam at 60fps.

Hi Gang

Thanks for reply

Since I applied the same config to single_cam and multi_cam, I can get totally the same images from these 2 apps. so this can fullfill my needs and I will close this question.

Regards

Zhou

Ok, so I will keep using DCC on single_cam

According to the config files:

Shutter time has been well calculated so exposures are the same.

Hi Zhou,

Xiaohan Zhou said:

Since I applied the same config to single_cam and multi_cam, I can get totally the same images from these 2 apps. so this can fullfill my needs and I will close this question.

The openvx kerenl of VISS takes input of (a) sensor raw images, (b) AE/AWB output (exposure time, WB gains, CCT etc.), (c) DCC tuning data.

If all of them are the same, the VISS output YUV image should be the same.

Xiaohan Zhou said:

Shutter time has been well calculated so exposures are the same.

In Brandon's experiment, we see the difference in raw, but did not see any other significant difference caused by VISS.