YOLO Task

Task for changing the bounding / anchor boxes from rectangles to ellipses (in train, val and detect) by Silvey Yu

Demo: On YouTube

How to run

1. Setup the virtual envrionment on your computer

pip3 install virtualenv
virtualenv test
source test/bin/activate

2. Clone this repository

cd test
git clone https://github.com/SilvesterYu/YOLO_Silvey_Task.git

3. Run setup

chmod -R 777 *.*
cd YOLO_Silvey_Task
chmod 0755 setup.sh
./setup.sh

⚠️ IMPORTANT: Make sure that you have torch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 exactly. You can check with

pip3 list

Testing

If "Descriptors cannot be created directly" error occurs, run:

pip install protobuf==3.20.1

1. Test Train

To test the train, go into yolov5copy directory

cd yolov5copy

Test train with

python train.py --data coco128.yaml --cfg yolov5s.yaml --weights '' --batch-size 128 --epochs 1

2. Test Val

To test the val, go into yolov5copy directory

cd yolov5copy

Test val with

python val.py --data coco128.yaml --weights yolov5s.pt --img 640 

3. Test Detection

To test the detection, go into yolov5copy directory

cd yolov5copy

Then, get an image from internet

wget "https://a-z-animals.com/media/2021/12/Best-farm-animals-cow.jpg"

Run detection using

python3 detect.py --weights yolov5s.pt --img 640 --conf 0.25 --source 'Best-farm-animals-cow.jpg' --save-txt