Problem with deserialize GenericProjectionFactor with python and Cauchy

[shteren1]

Description

When a GenericProjectionFactor is filled with serialized data and Cauchy robust loss it returns wrong error, using python wrapper.
I tested with Huber loss and it works fine. also notice that unweightenedError returns correct result.

Steps to reproduce

1. see the attached python code:

import gtsam
import numpy as np

point_noise = gtsam.noiseModel.Diagonal.Sigmas(np.ones(2))
point_noise = gtsam.noiseModel.Robust.Create(gtsam.noiseModel.mEstimator.Cauchy.Create(1), point_noise)
factor = gtsam.GenericProjectionFactorCal3_S2(np.ones(2), point_noise, 0, 1, gtsam.Cal3_S2())
values = gtsam.Values()
values.insert(0, gtsam.Pose3())
values.insert(1, np.array([2, 0, 15]))
#init new factor with something, since there is no empty constructor for GenerincProjectionFactor
factor1 = gtsam.GenericProjectionFactorCal3_S2(np.ones(2), point_noise, 0, 2, gtsam.Cal3_S2())
factor1.deserialize(factor.serialize())
print(f"original factor error: {factor.error(values)}, deserialized factor error: {factor1.error(values)}")

output:
original factor error: 0.24628311578204365, deserialized factor error: 0.003451520129521184

Expected behavior

The error returned will be similar

Environment

tested on ubuntu 18.04 with python 3.7 and gtsam pip package version 4.2a5

[ProfFan]

Hi @shteren1 I pushed the fix in the linked PR. Thanks for reporting :)