01. Background

Address, Decimation, Δt Event Representation (ADΔER)

The ADΔER (pronounced "adder") representation is inspired by the ASINT camera design by Singh et al. It aims to help us move away from thinking about video in terms of fixed sample rates and frames, and to provide a one-size-fits-all ("narrow waist") method for representing intensity information asynchronously.

Under the ASINT model, a pixel $\langle x,y\rangle$ continuously integrates light, firing an ADΔER event $\langle x,y,D,\Delta t\rangle$ when it accumulates $2^D$ intensity units (e.g., photons), where $D$ is a decimation threshold and $\Delta t$ is the time elapsed since the pixel last fired an event. we measure $t$ in clock “ticks,'' where the granularity of a clock tick length is user-adjustable. In a raw ADΔER stream, the events are time-ordered and spatially interleaved. An ADΔER event directly specifies an intensity, $I$, by $I \approx \frac{2^D}{\Delta t}$. The key insight of the ASINT model is the dynamic, pixelwise control of $D$. Lowering $D$ for a pixel will increase its event rate, while raising $D$ will decrease its event rate. With this multi-faceted $D$ control, we can ensure that pixel sensitivities are well-tuned to scene dynamics.

Practically speaking, it's most useful to think about ADΔER in reference to the source data type. In the context of framed video, ADΔER allows us to have multi-frame intensity averaging for stable (unchanging) regions of a scene. This can function both to denoise the video and enable higher dynamic range, all while preserving the temporal synchronicity of the source. See the info on simultaneous transcoding to quickly test this out!

What is "asynchronous video?"

I use the term "asynchronous video" to refer to any representation of spatial data (video) whose pixel value measurements are not temporally synchronous. That is, each pixel can express data at different points in time from any other pixel

What are event cameras?

Wikipedia has a decent summary. Basically, these are cameras whose pixels operate asynchronously, meaning that they directly capture asynchronous video. The dominant event camera technology used in the literature is the Dynamic Vision System (DVS).

What does this program do?

TODO