This package is used for object detection and segmentation

Dependences

First, create a conda env for yolov5 (we recommend python 3.10), e.g.

conda create -n yolo python=3.10
conda activate yolo

Then, in the folder of yolov5, install the requirements:

# in yolov5 path, e.g. src/vision/vision/yolov5
pip install -r requirements.txt

In order to use both python modules in ROS2 and those in conda env, you should change the conda env path mannully in the file src/vision/vision/obj_seg.py, src/vision/vision/obj_detect.py, etc.

# you should change the path here according to your conda env
sys.path.append(str(Path.home()) + '/miniforge3/envs/yolo/lib/python3.10/site-packages')

Train the network

Go to the path of yolov5

# make sure you have activated the conda env. 
# otherwise:
conda activate yolo

Then train the model.

python segment/train.py --img 960 --batch 8 --epochs 300 --data gazebo_seg.yaml --weights yolov5s-seg.pt --cache

the datasets and weights will be downloaded automatically.

• datasets in vision/datasets
• weights in yolov5/yolov5s-seg.pt

Remember to copy the training result (weights) from yolov5/runs/exp*/weights/best.pt to yolov5