search

implus / GFocal

Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection, NeurIPS2020
Apache License 2.0
587 stars 73 forks source link

Unstable training #15

Open harrylin-hyl opened 4 years ago

harrylin-hyl commented 4 years ago

image

I set the number of GPUs to 1, and then the predicted quality values of iou all become 0, making the training unable to proceed normally. Whether it can be improved without increasing the GPUs Thanks!