📝 Publications

Class-agnostic 3D Segmentation by Granularity-Consistent Automatic 2D Mask Tracking.
Juan Wang, Yasutomo Kawanishi, Tomo Miyazaki, Zhijie Wang, Shinichiro Omachi. -The research advances annotation-free 3D instance segmentation by tracking 2D masks across video and training progressively, turning noisy frame-wise pseudo labels into coherent 3D supervision for accurate, open-vocabulary results.

TAMC: Textual Alignment and Masked Consistency for Open-Vocabulary 3D Scene Understanding.
Juan Wang, Zhijie Wang, Tomo Miyazaki, Yaohou Fan, Shinichiro Omachi.

• The research improves 3D scene understanding through masked consistency training and pseudo-text labels, effectively addressing sparse point cloud processing and training-inference inconsistency issues.

An end-to-end deep neural network is proposed to detect aluminum profile defects \ Wang J, Meng Z H.

• Proposed a Deformable Feature Pyramid module to detect aluminum profile defects.