cutoff frequency and time constant warning

[gwgknudayanga]

Hi,

Following the v2e paper, to synthesize bright events i set the cutoff frequency value to 200 Hz. And the DVS timestamp resolution is set to 1ms.But then it gave the following warning. That means to synthesize events in bright mode (with 200Hz cutoff frequency) we always need to set the set the DVS timestamp resolution less than 1ms (in microsecond range) ? isn't it?

WARNING - Lowpass 3dB cutoff is f_3dB=200Hz (time constant tau=795.77us) with sample rate fs=1kHz (sample interval dt=1ms) , but this results in large IIR mixing factor eps = dt/tau = 1.257 > 0.3 (maxeps),

Thanks and Rgds, Udayanga

[tobidelbruck]

Sorry for very late reply, just came across this github notification. Yes, to properly filter the photoreceptor output, you must set a frame interval for upsampling via superslo mo that is sufficiently shorter than the lowpass filter time constant. I.e. if your cutoff frequency is 100Hz, then tau=1/(2pi100Hz)=1.6ms, so the frame interval has to be some small fraction of 1.6ms, e.g. 500us. It means a lot of computation unfortunately.

[gwgknudayanga]

Thanks for the reply. However, as i understood, to synthesize bright mode events (that is cutoff frequency is 200Hz as mentioned in v2e paper) we still need to set the cutoff frequency to 200Hz though the number of computations are high

[tobidelbruck]

Yes, if you want to model the motion blurring (which is pretty minimal except for fast moving stuff), then you need to lowpass filter the photoreceptor output. You can try turning it off and carefully inspecting the DVS output; maybe it is not important for your application. I recommend a native AER viewer that allows you to set the integration time of the DVS frames. Or else you can try outputting DVS frame videos with short and longer integration times.