A rotatable yolov3 model which can regress the angle of the bounding box
git clone https://github.com/woodsgao/rotatable_yolov3
cd rotatable_yolov3
pip install -r requirements.txt
Please organize your data in coco format(by default):
data/
<custom>/
images/
coco.json
train.json
val.json
You can use split_coco_json.py
from woodsgao/cv_utils
to split your coco.json
file into train.json
and val.json
python3 train.py data/<custom>
python3 -m torch.distributed.launch --nproc_per_node=<nproc> train.py data/<custom>
python3 test.py data/<custom>/val.json --weights weights.pth
python3 inference.py data/samples outputs --weights weights.pth
python3 export2caffe.py weights/best.pt -nc 21 -s 416 416