medeasolution / darknet-export-layers

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

tk darknet

this is a fork of alexey darknet tk version of yolos are in mycfg/ folder

Compile

Compile with make without modify anything in the makefile, it will use only CPU

make

Train

wget https://pjreddie.com/media/files/darknet53.conv.74 # pretrained weights
./darknet detector train data cfg darknet53.conv.74 -map

dont use pretrained weights with 1 channel input (for instance with yolo3_flir)

Test

./darknet detector demo data cfg weights test_video.mp4 -thresh 0.3

Export weights to tkDNN

./darknet export cfg weights out/path/

warning:

  • out/path should already be present and empty
  • debug is exported to "debug/" folder
  • if you need debug export of every layer ouput you MUST compile in CPU mode

About

License:Other


Languages

Language:C 65.2%Language:Cuda 15.6%Language:C++ 13.6%Language:Python 2.8%Language:CMake 1.4%Language:PowerShell 0.5%Language:Shell 0.5%Language:Makefile 0.3%Language:Batchfile 0.2%