Iterating to System Feasibility - 2024.2 English - XD100

Vitis Tutorials: AI Engine Development (XD100)

Document ID
XD100
Release Date
2024-12-06
Version
2024.2 English

Now having an accurate cost model of the Hough Transform from prototyping, you are in a position to explore alternate solutions via scaling. Here, consider “Image Partitioning” in addition to “Theta Partitioning” as a means to scale up the throughput. In this scheme, partition out portions of the images to 32-tile clusters, where each cluster is computing partial histograms for four $\theta$ values (as in the prototype). Using this approach, you can scale throughput linearly higher. As shown in the following diagram, you can achieve ~ 220 Mpps with ~275 tiles.

figure