Hello, I want to ask, should the words "NULL" and "loss_function" exchange?
They are in here: KeyFrameDatabase::optimize4DoFLoopPoseGraph.
problem.AddResidualBlock(cost_function, loss_function, euler_array[i-j],
t_array[i-j],
euler_array[i],
t_array[i]);
problem.AddResidualBlock(cost_function, NULL, euler_array[connected_index],
t_array[connected_index],
euler_array[i],
t_array[i]);
The technical report says sequential edges use 2norm-loss while loop edges use huber loss.
I tested the code on my own smartphone (after adaptions), and found that after the change the result became better during one test and remained similar during another test.
Hello, I want to ask, should the words "NULL" and "loss_function" exchange? They are in here: KeyFrameDatabase::optimize4DoFLoopPoseGraph. problem.AddResidualBlock(cost_function, loss_function, euler_array[i-j], t_array[i-j], euler_array[i], t_array[i]); problem.AddResidualBlock(cost_function, NULL, euler_array[connected_index], t_array[connected_index], euler_array[i], t_array[i]); The technical report says sequential edges use 2norm-loss while loop edges use huber loss. I tested the code on my own smartphone (after adaptions), and found that after the change the result became better during one test and remained similar during another test.