Darknet.jl
Wrapper for https://github.com/AlexeyAB/darknet based on pre-build binaries.
Current support:
- CPU-only with OpenMP via BinaryBuilder: https://github.com/JuliaPackaging/Yggdrasil/blob/master/D/Darknet
Platforms:
- i686-linux-gnu
- x86_64-linux-gnu
- aarch64-linux-gnu
- armv7l-linux-gnueabihf
- powerpc64le-linux-gnu
- i686-linux-musl
- x86_64-linux-musl
- aarch64-linux-musl
- armv7l-linux-musleabihf
- x86_64-apple-darwin14
- x86_64-unknown-freebsd11.1
- i686-w64-mingw32
- x86_64-w64-mingw32
Installation
Requires julia 1.3+. Install with Pkg, just like any other registered Julia package:
pkg> add Darknet # Press ']' to enter the Pkg REPL mode.Testing
using Darknet, Images
d = "/path/to/weights_and_config_files/"
weightsfile = "yolov3-tiny.weights"
cfgfile = "yolov3-tiny.cfg"
datafile = "coco.data"
imagefile = "/path/to/images/test.jpg"
net = Darknet.load_network(joinpath(d,cfgfile), joinpath(d,weightsfile),1)
meta = Darknet.get_metadata(joinpath(d,datafile));
Reading in an image from file:
# Read image using Darknet method
img = load(imagefile) #Image for plotting in julia purposes only (below)
img_d = Darknet.load_image_color(imagefile,0,0); #Darknet native way to read in image from file. Produces an image type with pointersor from an array in julia memory:
# Send image via an image in julia memory
img = convert(Array{Float32}, load(imagefile)) #Read in array via a julia method
img_d = Darknet.array_to_image(img) #Darknet image type with pointers to source dataor for looping through images from julia, avoid reallocation due to permuted dims:
img = convert(Array{Float32}, load(imagefile)) #Read in array via a julia method
# Darknet flips the first 2 dims of an image (cols,rows,colorchannels)
# so preallocate a permuted dims array to prevent reallocation in
if size(img,3) > 1 #if more than 1 color channel
img_permuted = Array{Float32}(undef,size(img,2),size(img,1),size(img,3))
else
img_permuted = Array{Float32}(undef,size(img,2),size(img,1))
end
img_d = Darknet.array_to_image(img,img_permuted) #Darknet image type with pointers to source dataRun detection
results = Darknet.detect(net,meta,img_d,thresh=0.1,nms=0.3)Preview result using Makie:
using Makie, GeometryTypes
scene = Scene(resolution = size(img'))
image!(scene,img',scale_plot = false)
for res in results
bbox = res[3]
poly!(scene,[Rectangle{Float32}(bbox[1]-(bbox[3]/2),bbox[2]-(bbox[4]/2),bbox[3],bbox[4])],color=RGBA(0,1,0,clamp(conf,0.05,0.5)))
end
scene