用合成数据学解耦表示，让其在真实图像上有效迁移。
Transferring disentangled representations: bridging the gap between synthetic and real images
- 用合成数据训练解耦表示，再迁移到真实图像。
- 迁移后仍保持一定解耦性，微调能进一步提升效果。
- 提出新可解释干预指标，评估编码质量。
构建有意义且高效的表征以分离数据生成机制的本质结构，在表征学习中至关重要。然而，由于生成因素相关、分辨率差异及真实标签获取受限，解耦表征学习在真实图像上的潜力尚未充分发挥。针对后者，本文探讨利用合成数据学习通用解耦表征并应用于真实数据的可行性，分析微调的影响以及解耦性质在迁移中的保留情况。我们进行了广泛的实证研究，并提出一种新的可解释干预度量指标，用于评估表征中因子编码的质量。结果表明，从合成数据到真实数据的解耦表示迁移是可行且有效的。
原文摘要 · Abstract (English)
Developing meaningful and efficient representations that separate the fundamental structure of the data generation mechanism is crucial in representation learning. However, Disentangled Representation Learning has not fully shown its potential on real images, because of correlated generative factors, their resolution and limited access to ground truth labels. Specifically on the latter, we investigate the possibility of leveraging synthetic data to learn general-purpose disentangled representations applicable to real data, discussing the effect of fine-tuning and what properties of disentanglement are preserved after the transfer. We provide an extensive empirical study to address these issues. In addition, we propose a new interpretable intervention-based metric, to measure the quality of factors encoding in the representation. Our results indicate that some level of disentanglement, transferring a representation from synthetic to real data, is possible and effective.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。