You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, there are multiple ways to define classes and their labels and two ways to process that information. It looks like the state is not in sync, which can lead to unexpected and tricky results. We need to implement a single source of truth and prevent situations that can lead to unexpected results.
Our methods must allow setting:
a class range without labels but with confidence constraints (leads to labels in class_0 format);
a label list defined with a file with confidence constraints (leads to the contiguous class range);
specific "label/class" pairs with confidence constraints and additional filtering options.
Our infrastructure, not nvinfer infrastructure, must handle all methods. Default processing must be implemented with our infrastructure, not nvinfer infrastructure.
The text was updated successfully, but these errors were encountered:
Currently, there are multiple ways to define classes and their labels and two ways to process that information. It looks like the state is not in sync, which can lead to unexpected and tricky results. We need to implement a single source of truth and prevent situations that can lead to unexpected results.
Our methods must allow setting:
class_0
format);Our infrastructure, not nvinfer infrastructure, must handle all methods. Default processing must be implemented with our infrastructure, not nvinfer infrastructure.
The text was updated successfully, but these errors were encountered: