param resume=True not work

THU-MIG / yolov10

YOLOv10: Real-Time End-to-End Object Detection

https://arxiv.org/abs/2405.14458

GNU Affero General Public License v3.0

9.37k stars 879 forks source link

param resume=True not work #409

Open micro-step opened 1 month ago

micro-step commented 1 month ago

yolo detect train data=coco.yaml model=runs/detect/train8/weights/last.pt epochs=80 batch=42 imgsz=832 device=0,1 resume=True

I find bug in check_resume(self, overrides) with file /home/zhanghao/ZK_goods_det/yolov10/ultralytics/engine/trainer.py. self.args = get_cfg(ckpt_args) overwrite param resume

babakbch commented 1 week ago

I have had this problem as well. I am trying to resume my training from a check point, but after much debugging I found out the optimizer state is stored in the checkpoint (self.optimizer.state_dict() in the save_model function) but is returning None. The key is stored but the state is returned as None.