Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration without Forgetting

Official Implementation, ICCV21

arxiv
Computer Vision Talks
poster | poster_slides | 5_min_video

Data

miniImageNet
here you can find all the splits and files for episodic training on mini-ImageNet (2.1 Gb)

pretrained_model on base 64 classes
there is also script to train the model from scratch
python mini_imgnet/run_pretrain_base.py

Run

set up all paths and run

python mini_imagenet/run_novel.py

Logs

set up params for logging of training, metrics and visdom in run_novel.py

files with harmonic mean and arithmetic mean in different spaces (generalized and not), for 5w1s
to save these files yourself turn write_in_file=True

details

pytorch 1.6
list of all packages that were installed but you do not need all of them [just in case]

cite

@inproceedings{kukleva2021lcwof,
author = {Kukleva, Anna and Kuehne, Hilde and Schiele, Bernt},
title = {Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration Without Forgetting},
booktitle = {ICCV},
year = {2021},
}