yolov5_ros

yolov5 with ROS

Environment

OS	Ubuntu 18.04
CUDA	10.2
Pytorch version	1.8
ROS version	Melodic
Language	Python3

Based on YOLOv5.

---

How to use

1. Locate weight files on weights folder, data.yaml files on data folder(maybe data.yaml file is not necessary).

2. Edit your detector.launch file.

<?xml version="1.0" encoding="UTF-8"?>
<launch>
    <arg name="weights_name"    default="yolov5s_door.pt" />
    <arg name="data_name"       default="door_handle.yaml" />

    <!-- Camera topic and weights, config arguments -->
    <arg name="image_topic"     default="/camera/color/image_raw" />
    <arg name="weights"         default="$(find yolov5_ros)/weights/$(arg weights_name)" />
    <arg name="data"            default="$(find yolov5_ros)/data/$(arg data_name)" />
    <arg name="width"           default="640" />
    <arg name="height"          default="480" />
    <arg name="conf_thres"      default="0.25" />

    <!-- Node -->
    <node name="detector" pkg="yolov5_ros" type="detect.py" output="screen" respawn="true">
        <param name="image_topic"   value="$(arg image_topic)" />
        <param name="weights"       value="$(arg weights)" />
        <param name="data"          value="$(arg data)" />
        <param name="width"         value="$(arg width)" />
        <param name="height"        value="$(arg height)" />
        <param name="conf_thres"    value="$(arg conf_thres)" />
    </node>
</launch>

Change arguments, "weights_name", "data_name" to yours. Also change "image_topic" too. Or you can just declare argument when you launch the launch file.

$ roslaunch yolov5_ros detector.launch weights_name:=${your weight file name} data_name:=${your yaml data file name} image_topic:=${image topic name}

Published topic

• /detected_img

sensor_msgs/Image

• /bounding_box_array

yolov5_ros/BoundingBoxes

header

bounding_boxes[]

string Class

float64 probability

int64 xmin

int64 ymin

int64 xmax

int64 ymax

TensorRT

If you want to use tensorrt model, file name "~~.engine", you have to check tensorrt version. If tensorrt version is mismatched, would face serialization error like this.

Serialization (Serialization assertion safeVersionRead == safeSerializationVersion failed.Version tag does not match.

So, it is recommended that make your own tenorrt model at your PC, with following this doc.