Is loss 4.x?

[luuuyi]

Hi, actually, follow your instruction, after 28000 iterations, the loss decrease to 4.x. So, my question is one model can be used correct, what loss should be decreased to? Or, I should reset the max iterations, 10000 or 200000? Thank you

[nyyznyyz1991]

for 20 layers, my loss value is 1.x and lfw test accuracy is 99.22%

[luuuyi]

how many iterations do you train? Can I have a look on your solver.prototxt?@nyyznyyz1991

[nyyznyyz1991]

i followed the author's solver,prototxt, didn't change anything. 28000 iters

base_lr: 0.1 display: 100 max_iter: 28000 lr_policy: "multistep" gamma: 0.1 momentum: 0.9 weight_decay: 0.0005 snapshot: 1000 snapshot_prefix: "result/sphereface_model" solver_mode: GPU device_id: 0 net: "code/sphereface_model.prototxt" train_state { level: 0 stage: "" } stepvalue: 16000 stepvalue: 24000 stepvalue: 28000

[luuuyi]

@nyyznyyz1991 ok, i see it. Some more question, have you follow the author's instruction one by one? Using the Matlab script to align the face, then train them? I ask this question because I just crop the face without aligning the face(using the five points). May be my problem happens at this?

[xhzcyc]

My LFW training loss is about 0.01

[Gerkam]

@luuuyi, I have the same results. I have it with align and without. Are you solve this problem?