Tiny YOLO v3 ZYNQ

/data

pre-trained layer weights for YOLOv3-tiny on COCO dataset coming from https://pjreddie.com/darknet/yolo/

'fp' stands for fixed point data 'bn' represents merging batch normalise into convolution

/hw

hardware IPs for each acceleration unit 'accumulation(acc)', 'convolution(conv)', 'max pooling(max)', 'upsample(upsamp)', 'yolo' '64' is used for 64b DMA transfer which is finally used 'prj' is the relevant SDK project to validate each IP

/model

models for Design Space Exploration (DSE)

/rpt

report for the project

/sw

purely software implementation of yolov3-tiny. It is a revised version of original https://github.com/pjreddie/darknet. File operations are removed for bare-mental running

/tools

some interesting helpers for pre-processing data

About

Implement Tiny YOLO v3 on ZYNQ

Languages

Language:C 31.2%Language:VHDL 30.7%Language:Coq 18.9%Language:C++ 9.3%Language:Objective-C 3.8%Language:Ada 3.0%Language:HTML 1.7%Language:SystemVerilog 0.5%Language:Verilog 0.4%Language:V 0.2%Language:Tcl 0.1%Language:Makefile 0.0%Language:JavaScript 0.0%Language:Scala 0.0%Language:Assembly 0.0%Language:Shell 0.0%Language:CartoCSS 0.0%Language:Batchfile 0.0%Language:Pascal 0.0%Language:Ruby 0.0%