-
Notifications
You must be signed in to change notification settings - Fork 50
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
人脸检测正确率波动 #8
Comments
正常,mini-batch 训练就是这样的 |
@Tim5Tang 你将layers文件夹下文件拷贝到caffe目录下编译出现的jfda_loss_param()未定义错误是怎么解决的? |
这个的话应该是你没有将损失函数层jfda,放到caffe 对应的位置 |
@Tim5Tang 谢谢你及时的回答,非常感谢。但我很确定按照作者提供的copy.sh将损失函数层jfda放到caffe对应的位置了 jfda_loss_param()确实未定义 因为根本搜索不到jfda_loss_param()的实现。可以在帮忙确定下吗 |
你那个proto文件中有没有jfda想关的参数,一般新加一个层需要几个部分,proto,头文件,layer都要看一下对不对 |
proto文件中是有用jfda_loss_param的 如下: |
jfda_loss_param { |
你好,非常感谢你的开源,这是我目前能跑完整的唯一一个mtcnn....
可能是我比较菜,但是我看了你的网络结构和lossfunction应该是我看到的所有mtcnn里面最清晰明了的了
我在训练R层的时候,发现pos_acc有较强的波动性不知道是什么原因,如下训练部分打印
I0602 11:04:08.421706 55441 solver.cpp:239] Iteration 13500 (39.3993 iter/s, 12.6906s/500 iters), loss = 1.55104
I0602 11:04:08.421797 55441 solver.cpp:258] Train net output #0: face_cls_loss = 0.0894116 (* 1 = 0.0894116 loss)
I0602 11:04:08.421815 55441 solver.cpp:258] Train net output #1: face_cls_neg_acc = 0.994792 (* 0.5 = 0.497396 loss)
I0602 11:04:08.421825 55441 solver.cpp:258] Train net output #2: face_cls_pos_acc = 0.953125 (* 1 = 0.953125 loss)
I0602 11:04:08.421844 55441 sgd_solver.cpp:112] Iteration 13500, lr = 0.005
I0602 11:04:21.311125 55441 solver.cpp:239] Iteration 14000 (38.7931 iter/s, 12.8889s/500 iters), loss = 1.55069
I0602 11:04:21.311210 55441 solver.cpp:258] Train net output #0: face_cls_loss = 0.153639 (* 1 = 0.153639 loss)
I0602 11:04:21.311228 55441 solver.cpp:258] Train net output #1: face_cls_neg_acc = 0.981771 (* 0.5 = 0.490885 loss)
I0602 11:04:21.311240 55441 solver.cpp:258] Train net output #2: face_cls_pos_acc = 0.882812 (* 1 = 0.882812 loss)
The text was updated successfully, but these errors were encountered: