inaturalist 2018
LearningImbalancedDatasetswith Label-Distribution-AwareMarginLoss
First, we propose a theoretically-principled label-distribution-aware margin (LDAM) loss motivated by minimizing a margin-based generalization bound.
Supplementary EscapingSaddlePointsforEffectiveGeneralizationon Class-ImbalancedData
We show results on CIFAR-100 LT with an imbalance factor (ฮฒ)of100and200.
Breaking Long-Tailed Learning Bottlenecks: A Controllable Paradigm with Hypernetwork-Generated Diverse Experts
We generate a set of diverse expert models via hypernetworks to cover all possible distribution scenarios, and optimize the model ensemble to adapt to any test distribution. Crucially, in any distribution scenario, we can flexibly output a dedicated model solution that matches the user's preference.
7 Appendix 7.1 Details of CutMix To sample the binary mask M, we first sample the bounding box coordinates B = (r x, r
Meanwhile, the confusion matrix is represented as two states, fixed and adaptive.