Distilling Virtual Examples for Long-tailed Recognition

ICCV 2021  ·  Yin-Yin He, Jianxin Wu, Xiu-Shen Wei ·

We tackle the long-tailed visual recognition problem from the knowledge distillation perspective by proposing a Distill the Virtual Examples (DiVE) method. Specifically, by treating the predictions of a teacher model as virtual examples, we prove that distilling from these virtual examples is equivalent to label distribution learning under certain constraints. We show that when the virtual example distribution becomes flatter than the original input distribution, the under-represented tail classes will receive significant improvements, which is crucial in long-tailed recognition. The proposed DiVE method can explicitly tune the virtual example distribution to become flat. Extensive experiments on three benchmark datasets, including the large-scale iNaturalist ones, justify that the proposed DiVE method can significantly outperform state-of-the-art methods. Furthermore, additional analyses and experiments verify the virtual example interpretation, and demonstrate the effectiveness of tailored designs in DiVE for long-tailed problems.

PDF Abstract ICCV 2021 PDF ICCV 2021 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Long-tail Learning ImageNet-LT DiVE Top-1 Accuracy 53.1 # 26
Long-tail Learning ImageNet-LT RIDE-DiVE Top-1 Accuracy 57.12 # 17
Long-tail Learning iNaturalist 2018 RIDE-DiVE Top-1 Accuracy 73.44% # 10
Long-tail Learning iNaturalist 2018 DiVE Top-1 Accuracy 71.71% # 14