Diego Rueda
Feb 6, 2015, 8:07:45 AM2/6/15
to caffe...@googlegroups.com
hi, I am trying to understand all the param transform possibilities in Caffe.
When I use mirror true, a mean file and crop size the loss converges gradually until i can get a 56% accuracy in validation, however when i use the scale 1/256 parameter the loss circles near the same value never decreasing even if i change the learning rate.
It is the same when i use it combined with the other param transformations or alone, can anyone help me understand what am I doing wrong?
And also any help i can get to top that 56% accuracy would be appreciated.
This is the dataset i am using http://www.robots.ox.ac.uk/~vgg/data/flowers/17/
When I use mirror true, a mean file and crop size the loss converges gradually until i can get a 56% accuracy in validation, however when i use the scale 1/256 parameter the loss circles near the same value never decreasing even if i change the learning rate.
It is the same when i use it combined with the other param transformations or alone, can anyone help me understand what am I doing wrong?
And also any help i can get to top that 56% accuracy would be appreciated.
This is the dataset i am using http://www.robots.ox.ac.uk/~vgg/data/flowers/17/
Reply all
Reply to author
Forward
0 new messages