Code, Algorithms and Probabilities

Part 2 ended on a diagnosis: the desktop and the Jetson TX2 disagree about which stage of the dense matcher is expensive, because they have different memory systems, and no single-machine implementation can be right for both. This post draws the conclusion: stop making one machine do both. The regular... [Read More]

In Part 1 I applied MASDA — max-sum loopy belief propagation for data association — to sparse stereo matching and measured what the one-to-one constraint buys against ground truth. It ended with a list of things that would improve the matcher. This post is what happened when I stopped matching... [Read More]

In Faster Data Association with Max-Sum Loopy Belief Propagation (MASDA) I derived MASDA for the tracking problem: associate measurements with objects, allowing for clutter and misdetection. Stereo matching has the same structure, so this post applies MASDA there and measures what it gains. [Read More]

This presents a data association algorithm I call MASDA (Max-Sum Algorithm Data Association) which relies on message passing in a factor graph. [Read More]

Gaussian Belief Propagation is a variant of Belief Propagation and used for inference on graphical models if the underlying distribution is described as a Gaussian. [Read More]