This repository is thought to facilitate the understanding of YOLOv5 by creating a simple file structure focused on the PyTorch implementation.
The sources I used while implementating are Ultralytics YOLOv5 for architecture details and Aladdin Persson YOLOv3 for the training-pipeline design.
git clone https://github.com/AlessandroMondin/YOLOV5m.git
cd YOLOv5m
pip install -r requirements.txt
To run experiments, your datasets need to comply with the below tree structure and the corresponing folder names.
yolov5m (this repository)
datasets
└── YOUR_DATASET
├── images
│ ├── train
│ └── val
└── labels
├── train
└── val
After loading Ultralytics COCO weights in my architecture, I've fine-tuned it on FLIR dataset for ~15 epoch and it reached ~0.82 of MAP50
