Reproduce “MORDA: A Synthetic Dataset to Facilitate Adaptation of Object Detectors to Unseen Real-target Domain While Preserving Performance on Real-source Domain” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hojun Lim, Heecheol Yoo, Jinwoo Lee, Seungmin Jeon, Hyeongseok Jeon
MORDA, a synthetic dataset developed by MORAI Inc., facilitates the adaptation of object detectors to unseen real-target domains while preserving performance on the real-source domain. This dataset, created by blending characteristics of nuScenes and South Korean environments, achieved a notable +22.17 mAP gain for 3D object detection on the unseen AI-Hub dataset and maintained or slightly improved performance on nuScenes.
Key insights
- Synthetically fusing relevant characteristics from both real-source and real-target domains can effectively bridge the domain gap for object detectors without requiring direct data collection from the target domain.
- This hybrid synthetic dataset acts as a regularization mechanism, enhancing the generalization capabilities of detectors and mitigating severe performance failures when deployed to unseen target domains.
- Careful and consistent replication of source domain sensor configurations, object categories, and labeling standards within the synthetic environment is crucial for maintaining or improving performance on the original source domain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.