Reproduce “Industrial Synthetic Segment Pre-training” 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
Shinichi Mae, Hirokatsu Kataoka, Ryousuke Yamada, Yoshihiro Fukuhara, Risa Shinoda, Christian Rupprecht
InsCore pre-trains instance segmentation models using automatically generated overlapping shapes with precise visible-region labels, rather than real images or manual annotations for pre-training. The method is fine-tuned on labeled industrial datasets. Across five domains, InsCore* scores 45.2 average mAP versus 45.0 for ImageNet-21k. Results vary by domain, and the ablation supports but does not isolate occlusion as the cause.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.