English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Synthetic-to-real transfer learning is a framework in which a syntheticallygenerated dataset is used to pre-train a model to improve its performance onreal vision tasks. The most significant advantage of using synthetic images isthat the ground-truth labels are automatically available, enabling unlimitedexpansion of the data size without human cost. However, synthetic data may havea huge domain gap, in which case increasing the data size does not improve theperformance. How can we know that? In this study, we derive a simple scalinglaw that predicts the performance from the amount of pre-training data. Byestimating the parameters of the law, we can judge whether we should increasethe data or change the setting of image synthesis. Further, we analyze thetheory of transfer learning by considering learning dynamics and confirm thatthe derived generalization bound is consistent with our empirical findings. Weempirically validated our scaling law on various experimental settings ofbenchmark tasks, model sizes, and complexities of synthetic images.

A Scaling Law for Synthetic-to-Real Transfer: How Much Is Your Pre-training Effective?