Reproduce “Dataset Enhancement with Instance-Level Augmentations” 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
Orest Kupyn, Christian Rupprecht
Researchers developed an instance-level data augmentation method utilizing pre-trained latent diffusion models to re-paint individual objects in images. This approach enhanced dataset diversity and model generalization across computer vision tasks, while also enabling data anonymization with minimal performance impact.
Key insights
- Leveraging pre-trained latent diffusion models allows for the injection of significant 'out-of-domain knowledge' and visual variability into datasets at the object instance level.
- The proposed method effectively preserves original image annotations and scene composition, avoiding the challenges of generating new annotations for synthetic data.
- Instance-level augmentation is effective for data anonymization, allowing replacement of sensitive objects with synthetic versions while maintaining overall dataset utility and strong privacy safeguards.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.