Horde Strategy for Class Incremental Learning with Repetition

Our paper "Incremental Learning with Repetition via Pseudo-Feature Projection" has been accepted at the 28th Computer Vision Winter Workshop (CVWW) at Graz.

This repository contains the official reference implementation. The code structure is based on FACIL.

Experiments

To run the experiments for our paper please run the corresponding scenario python files under "run_scripts". For example to reproduce the results for the base EFCIR scenario b, run:

python scenario_b_efcir.py

The CIFAR dataset for the experiment is automatically downloaded from torchvision.

Citation

If you find this works interesting, please cite us with the following

@inproceedings{tscheschner2025efcir,
               title={Incremental Learning with Repetition via Pseudo-Feature Projection},
               author={Tscheschner, Benedikt and Veas, Eduardo and Masana, Marc},
               journal={Computer Vision Winter Workshop},
               doi={10.3217/978-3-99161-022-9-004},
               year={2025}}