As I can understand, this parameter is related to sampling frequency of objects of the class. If this parameter = 0, objects of this class would not be sampled.
In argo2-3d-26class.py you define sample_groups as follows:
sample_group_1 = {k:1 for k in group1}
sample_group_2 = {k:2 for k in group2}
sample_group_3 = {k:2 for k in group3}
sample_group_4 = {k:1 for k in group4}
sample_group_5 = {k:2 for k in group5}
sample_group_6 = {k:2 for k in group6}
#merge all groups
sample_groups = {**sample_group_1, **sample_group_2, **sample_group_3, **sample_group_4, **sample_group_5, **sample_group_6}
sample_groups.update({'Wheelchair':0, 'Dog':0, 'Message_board_trailer':0})
How did you obtain these numbers(0, 1, or 2)?
In the FSDv2 paper you write about FSD:
the clustering part in instance segmentation voted centers necessitates pre-defined and handcrafted distance thresholds for each category, and it is non-trivial to find optimal values
Don't you think sample_groups values are encoding apriori information?
As I can understand, this parameter is related to sampling frequency of objects of the class. If this parameter = 0, objects of this class would not be sampled.
In argo2-3d-26class.py you define sample_groups as follows:
How did you obtain these numbers(0, 1, or 2)?
In the FSDv2 paper you write about FSD:
Don't you think sample_groups values are encoding apriori information?