There was a problem while training.

experiencor / keras-yolo2

Easy training on custom dataset. Various backends (MobileNet and SqueezeNet) supported. A YOLO demo to detect raccoon run entirely in brower is accessible at https://git.io/vF7vI (not on Windows).

MIT License

1.73k stars 784 forks source link

There was a problem while training. #323

Open emmawatsonll opened 6 years ago

emmawatsonll commented 6 years ago

default

Why is this happening? CRY = 0.00 , ANGRY = 0.00 ... Problem with Batch Size? Also, I wonder why it is early stop.

Help, please!

letilessa commented 6 years ago

Hi Emma, check the parameters in your early stop callback. Your training is being interrupted because the loss is too low and not showing any improvement according to what is defined in the callback.

early_stop = EarlyStopping(monitor='val_loss', min_delta=0.001, patience=3, mode='min', verbose=1)