Reproduce “TransCG: A Large-Scale Real-World Dataset for Transparent Object Depth Completion and a Grasping Baseline” 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

Hongjie Fang, Hao-Shu Fang, Sheng Xu, Cewu Lu

Researchers developed TransCG, a large-scale real-world dataset for transparent object depth completion, alongside DFNet, an efficient neural network that refines inaccurate depth maps. This work enables robust 6-DoF robotic grasping of transparent objects using standard RGB-D cameras.

Key insights

  • A semi-automatic data collection pipeline with object-level tracking and robot assistance can efficiently generate large-scale, high-quality real-world datasets for challenging perception tasks like transparent object depth completion.
  • Integrating raw sensor depth information throughout a U-Net architecture, combined with dense blocks and appropriate loss functions, leads to a highly efficient and accurate depth completion network.
  • Training on diverse, large-scale real-world datasets like TransCG allows models to generalize effectively to unseen transparent objects and bridge the synthetic-to-real domain gap, outperforming models trained purely on synthetic data.

Summary by alphaXiv.

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