How can I improve the learning of FCN32 for semantic segmentation?

Question

I have trained FCN32 for semantic segmentation from scratch for my data, and I got the following output:

As it can be seen, this is not a good learning curve showing an improper training on data. solver is as follows:

net: "train_val.prototxt"
#test_net: "val.prototxt"
test_iter: 5105 #736
# make test net, but don't invoke it from the solver itself
test_interval: 1000000 #20000
display: 50
average_loss: 50
lr_policy: "step" #"fixed"
stepsize: 50000 #+
gamma: 0.1  #+
# lr for unnormalized softmax
base_lr: 1e-10 
# high momentum
momentum: 0.99
# no gradient accumulation
iter_size: 1
max_iter: 600000
weight_decay: 0.0005
snapshot: 30000
snapshot_prefix: "snapshot/FCN32s_CNN1"
test_initialization: false
solver_mode: GPU

after changing the learning rate to 0.001, it became worse: I am wondering what can I do for improving the training? Thanks

score 0 · Answer 1 · answered Feb 20 '17 at 17:13

0

You can try varying the learning rate. Good values are normally something between 0.1 and 0.0001.

answered Feb 20 '17 at 17:13

Thomas Pinetz

6,948
2
27
46

Thanks for your suggestion, I edited the question, when I set the learning rate to 0.001, the learning curve becomes worse. – S.EB Feb 21 '17 at 02:55
The loss is actually better number wise. But your learning rate seems too high now. – Thomas Pinetz Feb 21 '17 at 05:56
Thanks Thomas for your suggestion – S.EB Feb 21 '17 at 07:33

How can I improve the learning of FCN32 for semantic segmentation?

1 Answers1