gain control of DAVIS 346 color

[Shijie Lin]

Hi,

I am using the DAVIS 346 colour. Usually, we can increase the gain and reduce the exposure time to reduce the motion blur for frame-based cameras. But I don't find the gain control of DAVIS, only the exposure time is found. Is there any way to increase/decrease the gain of the frame-based part in DAVIS?

Best,

Shijie Lin

[Luca Longinotti]

Hello, no this is not possible. The frame we output is the raw data
from the sensor with no post-processing, and the sensor itself has no
gain control, only the exposure affects how the frame capture behaves.
--
Luca Longinotti (llongi)

Head of Embedded Platforms
iniVation AG - https://inivation.com/
Zurich, Switzerland
Office: +41 44 500 32 14


-----Original Message-----
From: Shijie Lin <lsj...@connect.hku.hk>
To: davis-users <davis...@googlegroups.com>
Subject: [davis-users] gain control of DAVIS 346 color
Date: Tue, 25 Jan 2022 02:34:28 -0800

Hi,I am using the DAVIS 346 colour. Usually, we can increase the gain and

[Shijie Lin]

Dear Luca.

Thanks very much for the answer!

I have several questions that need your professional answer. 

Usually, frame-based cameras have a nonlinear radiometric response that can be described by the camera response function.

Is the frame output of DAVIS346 use a linear radiometric response? We use the images captured by DV to conduct a radiometric calibration, and we found the resulted camera response function is non-linear, more like a gamma mapping. So what happened inside this part?

It seems like the output of the frame from DAVIS240C ranges from 0-500, unlike output DAVIS346 which ranges from 0-255, so the camera response function of DAVIS240C is linear? (pls correct me, if I was wrong about anything)

Best,

Shijie Lin

[Luca Longinotti]

Both are technically linear, the value you get is simply the value of
the 10 bit ADC that measures pixel voltage to get an intensity measure.
DAVIS240C had non optimally set high/low references and so you only got
~9bits of actual range.
DAVIS346 cuts down always to 8 bits to fit into 8-bit integers,
previous experience showed us the extra 2 bits are not worth the
massive decrease in performance of having to use 16-bit integers, as
they provide almost no significant information.
So if you're calibrating based on light frequency on the Mono cameras
you should largely see values proportionally following the spectral
sensitivity here: https://inivation.com/faq/ (last question).
Hope this helps, I don't think we've ever done radiometric calibration
of a DAVIS, so please keep us informed of your progress. :-)

--
Luca Longinotti (llongi)

Head of Embedded Platforms
iniVation AG - https://inivation.com/
Zurich, Switzerland
Office: +41 44 500 32 14


-----Original Message-----
From: Shijie Lin <lsj...@connect.hku.hk>
To: davis-users <davis...@googlegroups.com>

Subject: Re: [davis-users] gain control of DAVIS 346 color
Date: Tue, 25 Jan 2022 18:26:03 -0800


Dear Luca.Thanks very much for the answer!

[Shijie Lin]

Dear Luca,

I can share with you some results captured by DAVIS 346 colour. I named the image with exposure time, and do the calibration using Matlab(https://www.mathworks.com/help/images/ref/camresponse.html). 

Seems like the response is not linear? I just don't know why I get this result. Maybe there is some tone mapping in DV?

[Tobi Delbruck (UZH-ETH)]

Thanks for sharing the results Shijie. I see your plot

But the xaxis is log and the y is linear, shouldn't you expect some exponential?

The DAVIS APS readout is rather simple 10 bit single slope ADC with a gray code capture of the digital value on crossing the swept reference value. The actual signal is difference between reset value readout and signal value readout. Generally the global shutter mode works best, at least is most supported. We have gotten Davis346 to have quite linear response over its limited DR of about 50dB, but the settings for the source follower, cascode, reference voltages, and various buffer biases need to be adjusted correctly. They may not be completely optimized in the defaults set by DV.

If you really care about this linearity, then I suggest using a chart like Edmund density step chart (https://www.edmundoptics.com/f/large-grayscale-target-399011f0/11434/) that allows capture of a wide range of values in single exposure. Of course you need to have uniform illumination of the chart too. Then you can play with the biases to optimize linearity most easily.

[Luca Longinotti]

If you're using the color, I don't think you should run calibration
against the actual color image, as that one is interpolated from the
Bayer pattern and you'll get computed values instead of the original
pixel response of the sensor.
The configuration 'aps/FrameMode', shown by default in the right-hand
menu, allows to switch to the 'Original' image as the sensor picks it
up, with no color interpolation (debayering) applied.

--
Luca Longinotti (llongi)

Head of Embedded Platforms
iniVation AG - https://inivation.com/
Zurich, Switzerland
Office: +41 44 500 32 14


-----Original Message-----
From: Shijie Lin <lsj...@connect.hku.hk>
To: davis-users <davis...@googlegroups.com>
Subject: Re: [davis-users] gain control of DAVIS 346 color

Date: Wed, 26 Jan 2022 09:19:18 -0800

Dear Luca,I can share with you some results captured by DAVIS 346 colour. I named

the image with exposure time, and do the calibration using

Matlab(https://www.mathworks.com/help/images/ref/camresponse.html).high
_light_original_mode.bmp