Hi!
I am trying to train faster-rcnn on my own dataset using the ZF model. I have trained stage 1 for 4000 iterations so far, but the loss is not changing much. The classification loss is staying around 0.2 and the regression loss is staying around 0.06. Will the losses converge after more iterations? I'm not sure if this is a problem with my data or if I haven't trained enough.
Also, what are good values for err_fg and err_bg? Mine stay at 0 and 1 respectively and I'm not sure what they're supposed to be.
Training : err_fg 0, err_bg 1, loss (cls 0.232 + reg 0.0576)
Testing : err_fg 0, err_bg 1, loss (cls 0.213 + reg 0.0804)
Hi! I am trying to train faster-rcnn on my own dataset using the ZF model. I have trained stage 1 for 4000 iterations so far, but the loss is not changing much. The classification loss is staying around 0.2 and the regression loss is staying around 0.06. Will the losses converge after more iterations? I'm not sure if this is a problem with my data or if I haven't trained enough.
Also, what are good values for err_fg and err_bg? Mine stay at 0 and 1 respectively and I'm not sure what they're supposed to be.
Thank you!