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

Deep learning in computer vision has achieved great success with the price oflarge-scale labeled training data. However, exhaustive data annotation isimpracticable for each task of all domains of interest, due to high labor costsand unguaranteed labeling accuracy. Besides, the uncontrollable data collectionprocess produces non-IID training and test data, where undesired duplicationmay exist. All these nuisances may hinder the verification of typical theoriesand exposure to new findings. To circumvent them, an alternative is to generatesynthetic data via 3D rendering with domain randomization. We in this work pushforward along this line by doing profound and extensive research on baresupervised learning and downstream domain adaptation. Specifically, under thewell-controlled, IID data setting enabled by 3D rendering, we systematicallyverify the typical, important learning insights, e.g., shortcut learning, anddiscover the new laws of various data regimes and network architectures ingeneralization. We further investigate the effect of image formation factors ongeneralization, e.g., object scale, material texture, illumination, cameraviewpoint, and background in a 3D scene. Moreover, we use thesimulation-to-reality adaptation as a downstream task for comparing thetransferability between synthetic and real data when used for pre-training,which demonstrates that synthetic data pre-training is also promising toimprove real test results. Lastly, to promote future research, we develop a newlarge-scale synthetic-to-real benchmark for image classification, termed S2RDA,which provides more significant challenges for transfer from simulation toreality. The code and datasets are available athttps://github.com/huitangtang/On_the_Utility_of_Synthetic_Data.

A New Benchmark: On the Utility of Synthetic Data with Blender for Bare Supervised Learning and Downstream Domain Adaptation