Towards Fine-grained Large Object Segmentation 1st Place Solution to 3D AI Challenge 2020 -- Instance Segmentation Track
arXiv:2009.04650
Abstract
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.
Technical Report
References in corpus (7)
- Res2Net: A New Multi-scale Backbone Architecture
- MMDetection: Open MMLab Detection Toolbox and Benchmark
- Deformable Convolutional Networks
- Hybrid Task Cascade for Instance Segmentation
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation
- Conditional Convolutions for Instance Segmentation
- 1st Place Solutions of Waymo Open Dataset Challenge 2020 -- 2D Object Detection Track