sazzadhrz / SAN

Official Implementation of the paper "Rethinking Task-Incremental Learning Baselines" accepted in the 26th International Conference on Pattern Recognition (ICPR 2022).

Rethinking Task-Incremental Learning Baselines

Official imlpementation of the paper [ICPR 2022]

Link to the Paper - arXiv

Prerequisites:

• Linux-64
• Python 3.9
• PyTorch 1.10.1
• CPU or NVIDIA GPU + CUDA10.2 CuDNN7.5

Installation

• Create a conda environment and install required packages:

conda create -n <env> python=3.9
conda activate <env>
pip install -r requirements.txt

Datasets

Download the Mini-imagenet and notMNIST datasets from Google Drive. Other datasets will be automatically downloaded.

Training

For training, run the following command.
python run.py -exp <experiment_id>

To manually input number of runs, epochs and learning rate, run the following command:
python run.py -exp <experiment_id> -r <n_runs> -e <n_epochs> -lr <learning_rate>

Test

For test, run the following command.
python test.py -exp <experiment_id>