taihulight / nms-speedup

A highly parallelized implementation of non-maximum suppression for object detection used for self-driving cars.

nms-speedup

A highly parallelized implementation of non-maximum suppression for object detection, used primarily for autonomous self-driving cars.

Installation

make

Run benchmark with bichen's data

python tests.py -f[v] [number of images]
full data benchmark expects an integer argument for the number of images to test, or -1 to test all images

Bichen's data available at https://www.dropbox.com/s/3p8n1q0ldz5v10a/data.tgz?dl=0. Download and extract

Run correctness on single file from bichen's data

python tests.py -c

Note: Performance results are very sensitive to load on the hive machines. If under load, OpenMP implementations severely cripple performance