Re-evaluating the Impact of Unseen-Class Unlabeled Data on SSL Model

This repository contains PyTorch implementation for our paper: Re-evaluating the Impact of Unseen-Class Unlabeled Data on SSL Model

Quick Start

1. varying the number of unseen-class examples

seen = unlabels_num * seen_ratio, unlabels_num, seen_ratio=0.2, change unseen_ratio

sh scripts/unseenc.sh 

2. varying the number of unseen-class categories

seen = unlabels_num * seen_ratio, unlabels_num, seen_ratio=0.2, unseen_ratio=0.2, change unseen_class_num

sh scripts/unseen_class_c.sh 

3. varying the index of unseen classes

seen = unlabels_num * seen_ratio, unlabels_num, seen_ratio=0.2, unseen_ratio=0.2, unseen_class_num=1, change unseen_class_index

sh scripts/unseen_index_c.sh 

4. varying the degrees of nearness in unseen classes

4.1 with the number of unseen-class categories

seen = unlabels_num * seen_ratio, unlabels_num=25000, seen_ratio=0.2, unseen_ratio=0.2, with MNIST, change unseen_class_num

sh scripts/unseen_near_c.sh 

4.1 with the number of unseen-class index

seen = unlabels_num * seen_ratio, unlabels_num=25000, seen_ratio=0.2, unseen_ratio=0.2, unseen_class_num=1, change unseen_class_index

sh scripts/unseen_near_index_c.sh 

5. varying the label distribution in unseen classes

seen = unlabels_num * seen_ratio, unlabels_num=25000, seen_ratio=0.2, unseen_ratio=0.4, unseen_class_num=5, change imb_factor

sh scripts/unseen_imbalance.sh