Training COCO dataset can't fit that total loss always higher than 70

lucasjinreal / yolov7_d2

🔥🔥🔥🔥 (Earlier YOLOv7 not official one) YOLO with Transformers and Instance Segmentation, with TensorRT acceleration! 🔥🔥🔥

GNU General Public License v3.0

3.13k stars 483 forks source link

Training COCO dataset can't fit that total loss always higher than 70 #169

Closed Louis8582 closed 2 years ago

Louis8582 commented 2 years ago

I used configs/coco/darknet53.yaml and didn't change anything. Only difference is that I trained with one GPU. My experiment's loss only decay to 70 then fluctuated. I saw the d2 default use 8 gpus, so do I need to change any setting in one gpu case?