CountButterfly

This task is based on yolov5 to produce butterfly detection
See the YOLOv5 for the reference

Install

a. Install conda and create virtual environment for this task

conda create -n countbutterfly python=3.7
conda activate countbutterfly
conda install pip

b. Clone yolov5 repo and install requirements.txt in a Python>=3.7.0 environment, including PyTorch>=1.7.

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

c. Clone this repo and put all the files into yolov5 folder.

git clone https://github.com/hhlim333/CountButterfly.git

d. Put butterfly.yaml into "data/" folder and (train,valid) folder into "data/image/" folder

1. Training Custom dataset (butterfly) for yolov5 model.

a. You can use your computer

cd yolov5
python train.py --img 250 --batch 14 --epochs 150 --data data/butterfly.yaml --weights yolov5s.pt --nosave --cache

b. google-colab
https://colab.research.google.com/drive/1LSNiUjIzOLYqoQ929K0qCcglM9ulgdK4?usp=sharing

After training, you can get the last.pt file for the trained model

2. Using trained model to produce butterfly detection for the youtube video.

Using counterbutterfly.py to show the result.

python countbutterfly.py