train_bottleneck both loss and val_loss are always nan

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)

MIT License

7.14k stars 3.44k forks source link

train_bottleneck both loss and val_loss are always nan #715

Open monicatao opened 4 years ago

monicatao commented 4 years ago

Hi everyone, I am training the model on my own dataset and I only have one class. anchors file is not changed. And I tried to train with and without pretrained.h5. But my loss and val_loss are always nan. Anyone have similar problem and the solution? Thank you! Screenshot 2020-08-10 at 21 05 31

ouarda253 commented 4 years ago

@monicatao i have the same probleme did you solve it ?

heshuju commented 3 years ago

I only have val_loss is nan, Because initial weights is not fit, and I use weights from Official website to solve it