Reproduce “Improving Long-Tail Detection via Layout-Instruction Conditioned Diffusion Synthesis” 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

Xinyu Zhang, Xiaodi Wang, Yanpeng Sun, Lingfeng Yang, Jingdong Wang, Lingqiao Liu

Tail-SD improves long-tail object detection by augmenting real training images with synthetic images from one diffusion model. The model accepts text instructions, object layouts, or both, using spatial guidance and semantic flexibility. On LVIS validation with CenterNet2, rare Box AP rises from 20.84 to 25.73 and rare Mask AP from 18.36 to 24.04, with a 20% synthetic-image ratio.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.