This project is an experimental implementation of the paper "Balance Calibration for Identifying Incorrect Classifications with Distinguishable Uncertainty" (under review in AISTATS 2022).
Override new subclass of torch.utils.data.Dataset
Implement init(), getitem() and len()
Save your code into balanced-TCP/helpers/MyDataset.py
Setup a new .py file into balanced-TCP/models/
Inherit torch.nn.Module
Implement init() and forward()
Follow the examples given in balanced-TCP/models/
Setup a new .yaml file into balanced-TCP/conf/
Follow the examples given in balanced-TCP/conf/
Revise balanced-TCP/RUNME.sh for detailed settings