YOLO-v1 implemented by pytorch.

This is just an experimental code to deepen the understanding of YOLO-v1. The overall design is based on YOLO-v1, but some structures have been changed, such as using deep separable convolutions and discarding fully connected layers. And i generated watermark data to verify the validity of the network.

---

train

easy to train

prepare training data, voc format

data/
    imgs/
        ...
    annotations/
        ...

Specify category name list and your hyper-parameter in train.py

nohup python train.py &

---

detect

python detect.py --input_path XXXXX

result

yolo-v2 implemented by myself on the way!!!

ref

my blog of yolo-v1

yolo-v1