SYNAuG: Exploiting synthetic data for data imbalance problems

Moon Ye-Bin, Nam Hyeon-Woo, Won-Seok Choi, Nayeong Kim, Suha Kwak, Tae-Hyun Oh

Abstract: Highlights•We propose SYNAuG for investigating and exploiting the efficacy of synthetic data.•We found that at least a few real samples are essential, and a domain gap exists.•Various experiments verify the effectiveness of synthetic data in imbalance problems.•Mixup and classifier finetune help reduce the domain gap between real and synthetic.