This technical report introduces our solutions of Team 'FineGrainedSeg' for Instance Segmentation track in 3D AI Challenge 2020. In order to handle extremely large objects in 3D-FUTURE, we adopt PointRend as our basic framework, which outputs more fine-grained masks compared to HTC and SOLOv2. Our final submission is an ensemble of 5 PointRend models, which achieves the 1st place on both validation and test leaderboards. The code is available at https://github.com/zehuichen123/3DFuture_ins_seg.

翻译：本技术报告介绍了团队“FineGrainedSeg”在3D AI挑战赛2020实例分割赛道中的解决方案。为处理3D-FUTURE数据集中极端庞大的物体，我们采用PointRend作为基础框架，该框架相比HTC和SOLOv2能够输出更加精细的掩膜。最终提交的成果为5个PointRend模型的集成结果，在验证集和测试集排行榜上均取得第一名。相关代码已开源至https://github.com/zehuichen123/3DFuture_ins_seg。