TY - GEN
T1 - Impact of Base Dataset Design on Few-Shot Image Classification
AU - Sbai, Othman
AU - Couprie, Camille
AU - Aubry, Mathieu
N1 - Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
PY - 2020/1/1
Y1 - 2020/1/1
N2 - The quality and generality of deep image features is crucially determined by the data they have been trained on, but little is known about this often overlooked effect. In this paper, we systematically study the effect of variations in the training data by evaluating deep features trained on different image sets in a few-shot classification setting. The experimental protocol we define allows to explore key practical questions. What is the influence of the similarity between base and test classes? Given a fixed annotation budget, what is the optimal trade-off between the number of images per class and the number of classes? Given a fixed dataset, can features be improved by splitting or combining different classes? Should simple or diverse classes be annotated? In a wide range of experiments, we provide clear answers to these questions on the miniImageNet, ImageNet and CUB-200 benchmarks. We also show how the base dataset design can improve performance in few-shot classification more drastically than replacing a simple baseline by an advanced state of the art algorithm.
AB - The quality and generality of deep image features is crucially determined by the data they have been trained on, but little is known about this often overlooked effect. In this paper, we systematically study the effect of variations in the training data by evaluating deep features trained on different image sets in a few-shot classification setting. The experimental protocol we define allows to explore key practical questions. What is the influence of the similarity between base and test classes? Given a fixed annotation budget, what is the optimal trade-off between the number of images per class and the number of classes? Given a fixed dataset, can features be improved by splitting or combining different classes? Should simple or diverse classes be annotated? In a wide range of experiments, we provide clear answers to these questions on the miniImageNet, ImageNet and CUB-200 benchmarks. We also show how the base dataset design can improve performance in few-shot classification more drastically than replacing a simple baseline by an advanced state of the art algorithm.
KW - Dataset labeling
KW - Few-shot classification
KW - Meta-learning
KW - Weakly-supervised learning
UR - https://www.scopus.com/pages/publications/85092896429
U2 - 10.1007/978-3-030-58517-4_35
DO - 10.1007/978-3-030-58517-4_35
M3 - Conference contribution
AN - SCOPUS:85092896429
SN - 9783030585167
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 597
EP - 613
BT - Computer Vision – ECCV 2020 - 16th European Conference, Proceedings
A2 - Vedaldi, Andrea
A2 - Bischof, Horst
A2 - Brox, Thomas
A2 - Frahm, Jan-Michael
PB - Springer Science and Business Media Deutschland GmbH
T2 - 16th European Conference on Computer Vision, ECCV 2020
Y2 - 23 August 2020 through 28 August 2020
ER -