Kang Cheng presents Class-Balanced Loss Based on Effective Number of Samples

On 2021-02-25 11:00:00 at https://feectu.zoom.us/j/97810240340
Reading group on the work "Class-Balanced Loss Based on Effective Number of
Samples" (CVPR 2019) by Yin Cui, Menglin Jia, Tsung-Yi Lin, Yang Song, Serge
Belongie presented by Kang Cheng.

Paper abstract:
With the rapid increase of large-scale, real-world datasets, it becomes
to address the problem of longtailed data distribution (i.e., a few classes
account for most of the data, while most classes are under-represented).
Existing solutions typically adopt class re-balancing strategies such as
re-sampling and re-weighting based on the number of observations for each
In this work, we argue that as the number of samples increases, the additional
benefit of a newly added data point will diminish. We introduce a novel
theoretical framework to measure data overlap by associating with each sample a
small neighboring region rather than a single point. The effective number of
samples is defined as the volume of samples and can be calculated by a simple
formula (1−βn)/(1−β), where n is the number of samples and β ∈ [0, 1)
is a hyperparameter. We design a re-weighting scheme that uses the effective
number of samples for each class to re-balance the loss, thereby yielding a
class-balanced loss. Comprehensive experiments are conducted on artificially
induced long-tailed CIFAR datasets and large-scale datasets including ImageNet
and iNaturalist. Our results show that when trained with the proposed
class-balanced loss, the network is able to achieve significant performance
gains on long-tailed datasets.

Paper URL:

Online meeting: https://feectu.zoom.us/j/97810240340

Instructions for participants: The reading group studies the literature in the
field of pattern recognition and computer vision. At each meeting one or more
papers are prepared for presentation by a single person, the presenter. The
meetings are open to anyone, disregarding their background. It is assumed that
everyone attending the reading group has, at least briefly, read the paper –
not necessarily understanding everything. Attendants should preferably send
questions about the unclear parts to the speaker at least one day in advance.
During the presentation we aim to have a fruitful discussion, a critical
analysis of the paper, as well as brainstorming for creative extensions.

See the page of reading groups
Responsible person: Petr Pošík