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:这份技术报告介绍了我们“FineGraineedSeg”团队在3D AI 2020 挑战中为事件分解轨道采用的解决办法。为了处理3D-Future中的极大型物体,我们采用PointRend作为我们的基本框架,与HTC和SOLOv2相比,PointRend提供了更精细的面罩。我们最后提交的5点Rind模型合在一起,在验证和测试首选板上都达到了第一位置。该代码可在https://github.com/zehuichen123/3DFuture_ins_seg上查阅。