Dog Cat Detection with YOLOv5

Simple project that applied YOLOv5 in detection of dogs and cats

Environment and Dependencies

• Python 3.10.6

Install requirements

pip install -r requirements.txt

Data

• Data used for this project can be found here.
• Dataset creation: Refer to YOLOv5 Train Custom Data for more information. In short, labels and bouding boxes were converted in to .txt format as follow: class x_center y_center width height
• Data config for training: Standard YOLOv5 YAML format. Refer to dataset.yaml for details.

path: ../data/  # dataset root dir
train: train/images  # train images (relative to 'path') 80% images
val: valid/images  # val images (relative to 'path') 20% images
test:  # test images (optional)

# Classes (2 classes)
names:
  0: cat
  1: dog

Training

The model was trained with 10 epochs and standard COCO128 configuration (batch-size, image size and pretrained --weights yolov5s.pt)

python train.py --img 640 --batch 16 --epochs 10 --data ../dataset.yaml --weights yolov5s.pt --device 0

Training results on valid set: Metrics:

Inference with detect.py

detect.py runs inference on a variety of sources.

python detect.py --weights [path-to-model] --source [path-to-file] 

Input Image	Result Image

Reference

• YOLOv5