Some questions about the relationship between the contrast sensitivity and event probability

[Li]

Hello, I have some questions about the relationship between the contrast sensitivity and the probability of an event in the DAVIS.
For example, in DVXplorer, a contrast sensitivity of 13% means 50% event probability,  and 27.5% means 99.9% . Does that mean that if I set a contrast sensitivity of 7% it means that there is less than 50% probability of the events?Is this an intrinsic feature of the hardware or is the default setting, but I can modify the contrast sensitivity at 50% probability of events? Because I've found in other brands of event cameras, such as Prophesee, that their definition of contrast threshold is the contrast ratio of an event with a 50% probability.

And  why different dynamic ranges correspond to different relationships between those two?Are the settings the same for both tests?
Could you please explain the relationship between the contrast sensitivity and the probability of the event,?

thank you very much!

[li3.che...@gmail.com]

Hi, in an ideal noise free scenario, an input that just reaches the contrast threshold would produce an event. Now we consider noise, which typically has a symmetrical probability distribution centered at the ground truth. Therefore, if the ground truth input reaches the contrast threshold, there's a 50% chance it can produce an event. That's why the 50% event probability is used to measure the contrast threshold. A higher event probabiilty such as 99.9% is just a more strict measurement condition, and doesn't necessarily mean a different hardware configuration, it simply means that when the ground truth input has a 27.5% contrast, 99.9% of its probable detected value taking noise into account would exceed the configured event threshold, which could very well be the same configuration as measued using the 50% event probablity.
Different dynamic range corresponds to different contrast sensitivity because the contrast sensitivity degrades under low light, when noise starts to bury the input signal. Therefore, if the input signal has a high contrast such as 80%, then it can still produce events with 50% chance under 0.3 lux illumination. But if the input signal has a smaller contrast, such as 27.5%, it will have much lower chance to produce events under 0.3 lux.

The above is an explain on how to interprete the specs, I have not performed these measurements myself to verify these numbers. Hope this helps.

Chenghan

[Tobi Delbruck (INI)]

Thanks Chenghan.

Also this 2011 ISCAS paper from Christoph Posch explains and shows data
nicely: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5937877
under a few illumination levels. And lately DVS papers have
characterized NCT like this too, e.g. this 2023 Sony paper:
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10067566


Tobi