Should We Always Train Models on Fine-Grained Classes?

This repository is the official implementation of "Should We Always Train Models on Fine-Grained Classes?.

Requirements

To create and activate the Conda requirements from the provided .txt file:

1. Create the environment:

   conda create -n subclasses -f requirements.txt

2. Activate the environment:

   conda activate subclasses

Training

Training and evaluation is available in the coarse_vs_fine_experiments.ipynb for Fig. 2 and 3