island_loss

Island loss implement by tensorflow Refer to the paper "Island Loss for Learning Discriminative Features in Facial Expression" The basic code is referred to https://github.com/EncodeTS/TensorFlow_Center_Loss, besides I read the relative part in https://github.com/davidsandberg/facenet/blob/master/src/facenet.py#L76-L88, if i use the latter one which use tensorflow auto grad computation,the center points in tensorboard will appear only one point while i use the tf.control_depencies[update_op] in the primer ,the center points will appear several points correspond to the center points' numbers.The picture will be attached latter. About the center point initializer,the center loss uses zeros_initializer,but it will cause grad NAN in island loss, so gaussain initializer instead of the original one. About the converage,more tests need to be tried.