IEF (Iterative Error Feedback): Iterative feedback
Objective: To make more accurate prediction of parameters
research ideas:
(1) in an iterative manner to update the parameters you want to forecast
(2) the output of this error is the error ----- The shape variable of the parameter
(3) after multiple iterations is the parameter that needs to be predicted
for i in np.arange(self.num_stage):
print('Iteration %d' % i)
# theta_prev 表示的是需要预测的参数
state = tf.concat([self.img_feat, theta_prev], 1)
if i == 0:
# delta_theta 表示的是预测参数的形变量
delta_theta, threeD_var = threed_enc_fn(
state,
num_output=self.total_params,
reuse=False)
else:
delta_theta, _ = threed_enc_fn(
state, num_output=self.total_params, reuse=True)
"""
这里面忽略了你想要对于预测参数的操作
"""
# (形成迭代误差反馈了) 通过误差来更新需要预测的参数
theta_here = theta_prev + delta_theta
# Finally update to end iteration.---> theta_here update the theta
# 迭代误差反馈的入口
theta_prev = theta_here
The overall idea is to update the desired parameters through errors-through iterative methods-
welcome to leave a message