Box_detection_Yolov3_Darknet

Yolo v3 Darknet for box detection in gazebo. This package provides dataset and training notebooks. As well as trained weights

Reference : Here

Note : For unknown reasons , yolo v3 tiny weights don't show any predictions when used with darknet ( AlexeyAB ) but Yolo v3 weights work fine.

Darknet ROS Installation

darknet_ros

cd ~/catkin_ws/src

git clone --recursive https://github.com/leggedrobotics/darknet_ros.git

cd ~/catkin_ws

catkin_make -DCMAKE_BUILD_TYPE=Release

For testing performance of trained weights :

Yolo v3 Darknet ( AlexeyAB )

• Download weights : Here
• Download sample test images : Here

cd ~
git clone https://github.com/AlexeyAB/darknet.git
cd darknet

( Make sure you have downloaded necessary libraries , please refer to these guides 1 and 2 )

Make changes in Makefile and save them using gedit or code (vs code)

code Makefile

For CPU build , set following parameters in Makefile :

CUDNN=0
CUDNN_HALF=0
OPENCV=1
AVX=1
OPENMP=1
LIBSO=1  
ZED_CAMERA=0
ZED_CAMERA_v2_8=0

Save the edited Makefile

make

Now copy obj.data , obj.names and yolov3_training.cfg files from ~/Object_follower_UR5/src/yolov3/cfg folder to cfg folder of darknet directory. Also place weights file in darknet directory Now place any test image ( for example here it is two_boxes.png ) in darknet directory and run following command :

cd ~/darknet
./darknet detector test cfg/obj.data cfg/yolov3_training.cfg yolov3_training.weights two_boxes.png

Watch the predictions of model! (Task 3 - Object Detection)