search

wuqianliang / deepid2_caffe

deepid2 face verification base on caffe.
15 stars 8 forks source link

> Why this error was thrown when I run this net? #7

Open Zoejkr opened 5 years ago

Zoejkr commented 5 years ago

Why this error was thrown when I run this net? registry.count(type) == 1 (0 vs. 1) Unknown layer type: Identity2Verify

You should merge deepid2_caffe/caffe_master/src/caffe/layers/ into your caffe dir.

_Originally posted by @wuqianliang in https://github.com/wuqianliang/deepid2_caffe/issues/6#issuecomment-470001189_

Thanks，that problem has been solved. But during trianning the deepid2 net, the sim_loss was always 0, and the last accuracy was about 90.3%. So why the sim_loss was 0?

Yodamt commented 5 years ago

Why this error was thrown when I run this net? registry.count(type) == 1 (0 vs. 1) Unknown layer type: Identity2Verify

You should merge deepid2_caffe/caffe_master/src/caffe/layers/ into your caffe dir.

_Originally posted by @wuqianliang in #6 (comment)_

Thanks，that problem has been solved. But during trianning the deepid2 net, the sim_loss was always 0, and the last accuracy was about 90.3%. So why the sim_loss was 0?

i keep training the lfw-aligned set ,however accuracy keeps going down along 65percent , could u teel me what training-set or param for fullconnect layer are you using ? thanks a lot

wuqianliang commented 5 years ago

Why this error was thrown when I run this net? registry.count(type) == 1 (0 vs. 1) Unknown layer type: Identity2Verify

You should merge deepid2_caffe/caffe_master/src/caffe/layers/ into your caffe dir. _Originally posted by @wuqianliang in #6 (comment)_ Thanks，that problem has been solved. But during trianning the deepid2 net, the sim_loss was always 0, and the last accuracy was about 90.3%. So why the sim_loss was 0?

i keep training the lfw-aligned set ,however accuracy keeps going down along 65percent , could u teel me what training-set or param for fullconnect layer are you using ? thanks a lot

Actually, I use the MS-celeB-1M datasets as training-set while lfw-aligned as testind-set afterwards, and finally use the sphereface as my model; I found that the sphereface model much strongger than deepid model series