Towards Fine-grained Large Object Segmentation 1st Place Solution to 3D AI Challenge 2020 -- Instance Segmentation Track
This work addresses instance segmentation for large objects in 3D scenes, but it is incremental as it applies an existing method to a specific dataset.
The paper tackled instance segmentation of extremely large objects in the 3D-FUTURE dataset by using PointRend as the basic framework, achieving first place in the 3D AI Challenge 2020 with an ensemble of 5 models.
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.