search

weiliu89 / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
4.77k stars 1.68k forks source link

Does more iteration improve detection accuracy? #167

Open 1292765944 opened 8 years ago

1292765944 commented 8 years ago

I notice that in the solver.prototxt the max_iter is only 60000, How do you select the iterations. I notice that when I train the model, the loss is not stably decreased, So when should we stop training? Does more iterations improve detection accuracy? Thank you!

ck196 commented 8 years ago

You can try it. You can plot loss graph to decide when should stop training. I often stop training when loss does not decreased after 10000 iteration. If your training has loss is not stably decreased, something was wrong. You should check your model, data...

sunwooOh commented 8 years ago

You can change the parameters(e.g. max iterations, learning rate) in $CAFFE_ROOT/examples/ssd/ssd_pascal.py. It's in the solver_params dict, if you're still looking for it.