supitalp
commented
5 years ago Hi, The relationship between the biases and the actual contrast threshold is complex, and as far as I know there exists no such simple mapping, as this does not only depend on the ONBn and OFFBn biases, but also many additional ones such as the refractory periods, etc. (see more info here: https://inivation.com/support/hardware/biasing/).
If you have access to a DAVIS sensor, you may try to estimate the contrast thresholds using the pixel intensities from consecutive frames (see this paper for example). However, their value will jitter a lot.
That being said, the naive reconstruction algorithm you suggest will unfortunately not work (even if you estimate the contrast thresholds roughly): it will diverge quickly due to noise. If you are interested in image reconstruction, you may try our reconstruction code.
The threshold of the DVS for event generation is determined by the setting of the biasing currents in the circuit. In this driver, the same is achieved by changing the sliders corresponding to ONBn and OFFBn (both course and fine). Is there a way to translate the numbers in the slider setting to the actual value of threshold for use in a reconstruction algorithm?
Suppose events happen at t+1, I can naively reconstruct the image at t+1 by adding threshold*event_polarity to the Frame image at t. For this I would need to know the correspondence between the slider setting and the threshold value.