Query about the categorical Scores

[soham-joshi]

Hi @cshizhe , since the code for the categorical evaluation is not public. I wrote a script for evaluating the model in 'Easy', 'Hard', 'View-Dependent', and 'View-Independent' categories. I re-trained the baseline model and got an accuracy of 72.55 on the SR3D dataset. Following are the categorical scores of the same re-trained baseline model.
{'Overall': 72.55443980593479, easy': 74.20756234915527, 'hard': 68.67447129909365, 'vd': 57.446808510638306, 'vid': 73.2237539766702}

There's a large difference between the scores re-evaluated and the reported score (paper: https://arxiv.org/abs/2211.09646) in the View-Dependent Category. Just wanted you to clarify if it is right or what could have gone wrong.

Thanks,
Soham Joshi