Kezunlin's Blog

Live and Learn

0%

compile darknet on ubuntu 16.04

Posted on Edited on In

Series

Guide

Darknet is an open source neural network framework written in C and CUDA. It is fast, easy to install, and supports CPU and GPU computation.

optional dependencies:

  • opencv
  • gpu (CUDA/CUDNN)

repo intro

see darknet
This repository supports:

  • both Windows(VS 2015 (v140)) and Linux (GCC>=4.9)
  • CPU and GPU (with CC >= 3.0)
  • both OpenCV 2.x.x and OpenCV <= 3.4.0 (3.4.1 and higher isn’t supported)
  • both cuDNN v5-v7
  • CUDA >= 7.5
  • also create SO-library on Linux and DLL-library on Windows

compile

my system requirements (same as caffe on ubuntu 16.04)

  • ubuntu 16.04
  • GeForce 1060 (6G)
  • cuda: 9.2
  • cudnn: 7.1.4
  • opencv: 3.3.0
  • openmp

commands

1
2
3
4
git clone https://github.com/AlexeyAB/darknet.git

cd darknet
vim Makefile

options:

GPU=1
CUDNN=1
CUDNN_HALF=0
OPENCV=0
AVX=0
OPENMP=1
LIBSO=1

now compile with 

1
make -j8

OK. now we have darknet, darknet.so and uselib which use darknet.so

fix error

CUDNN error:

Try to load cfg: ./ped.cfg, weights: ./ped.weights, clear = 0 
   layer   filters  size/strd(dil)      input                output
   0 
 cuDNN status Error in: file: ./src/convolutional_layer.c : () : line: 301 : build time: Sep 24 2019 - 10:28:26 
cuDNN Error: CUDNN_STATUS_BAD_PARAM

fix

Makefile options:

CUDNN=0

and compile again

1
make -j8

see here

pretrained models

dataset

  • coco: 80 classes
  • voc: 20 classes

model

  • yolov3: 0-106 layers

  • yolov2: 0-31 layers

  • yolov3.cfg (236 MB COCO Yolo v3) - requires 4 GB GPU-RAM: yolov3.weights

  • yolov2.cfg (194 MB COCO Yolo v2) - requires 4 GB GPU-RAM: yolov2.weights

demo

yolov2

1
sh ./image_yolov2.sh

or 

1
./darknet detector test ./cfg/coco.data ./cfg/yolov2.cfg ./yolov2.weights data/dog.jpg -i 0 -thresh 0.2

output 

 29 conv   1024  3 x 3 / 1    13 x  13 x1280   ->    13 x  13 x1024 3.987 BF
  30 conv    425  1 x 1 / 1    13 x  13 x1024   ->    13 x  13 x 425 0.147 BF
  31 detection
mask_scale: Using default '1.000000'
Total BFLOPS 29.475 
Loading weights from ./yolov2.weights...
 seen 32 
Done!
data/dog.jpg: Predicted in 0.000000 milli-seconds.
dog: 79%
bicycle: 84%
truck: 77%
Not compiled with OpenCV, saving to predictions.png instead

view results

1
eog prediction.png

yolov2 result

yolov3

1
sh ./image_yolov3.sh

or 

1
./darknet detector test ./cfg/coco.data ./cfg/yolov3.cfg ./yolov3.weights data/dog.jpg -i 0 -thresh 0.25

output 

104 conv    256  3 x 3 / 1    52 x  52 x 128   ->    52 x  52 x 256 1.595 BF
 105 conv    255  1 x 1 / 1    52 x  52 x 256   ->    52 x  52 x 255 0.353 BF
 106 yolo
Total BFLOPS 65.864 
Loading weights from ./yolov3.weights...
 seen 64 
Done!
data/dog.jpg: Predicted in 0.000000 milli-seconds.
bicycle: 99%
dog: 100%
truck: 93%
Not compiled with OpenCV, saving to predictions.png instead

view results

1
eog prediction.png

yolov3 result

uselib

1
./uselib data/dog.jpg

output 

100 conv    256  3 x 3 / 1    52 x  52 x 128   ->    52 x  52 x 256 1.595 BF
 101 conv    128  1 x 1 / 1    52 x  52 x 256   ->    52 x  52 x 128 0.177 BF
 102 conv    256  3 x 3 / 1    52 x  52 x 128   ->    52 x  52 x 256 1.595 BF
 103 conv    128  1 x 1 / 1    52 x  52 x 256   ->    52 x  52 x 128 0.177 BF
 104 conv    256  3 x 3 / 1    52 x  52 x 128   ->    52 x  52 x 256 1.595 BF
 105 conv    255  1 x 1 / 1    52 x  52 x 256   ->    52 x  52 x 255 0.353 BF
 106 yolo
Total BFLOPS 65.864 
Loading weights from yolov3.weights...
 seen 64 
Done!
object names loaded 
input image or video filename: dog - obj_id = 16,  x = 123, y = 223, w = 196, h = 319, prob = 0.998
truck - obj_id = 7,  x = 474, y = 87, w = 216, h = 78, prob = 0.931
bicycle - obj_id = 1,  x = 117, y = 124, w = 451, h = 308, prob = 0.99

darknet.py

1
python darknet.py

Reference

History

  • 20181031: created.