fcaf3d

FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection

FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection

Abstract

Recently, promising applications in robotics and augmented reality have attracted considerable attention to 3D object detection from point clouds. In this paper, we present FCAF3D --- a first-in-class fully convolutional anchor-free indoor 3D object detection method. It is a simple yet effective method that uses a voxel representation of a point cloud and processes voxels with sparse convolutions. FCAF3D can handle large-scale scenes with minimal runtime through a single fully convolutional feed-forward pass. Existing 3D object detection methods make prior assumptions on the geometry of objects, and we argue that it limits their generalization ability. To eliminate prior assumptions, we propose a novel parametrization of oriented bounding boxes that allows obtaining better results in a purely data-driven way. The proposed method achieves state-of-the-art 3D object detection results in terms of mAP@0.5 on ScanNet V2 (+4.5), SUN RGB-D (+3.5), and S3DIS (+20.5) datasets.

Introduction

We implement FCAF3D and provide the result and checkpoints on the ScanNet dataset. Max and mean metrics are copied from the paper and ours is for provided checkpoint. Mean value is averaged across 5 train runs followed by 5 test runs. Inference time is given for a single NVidia GTX1080ti GPU. All models are trained on 2 GPUs.

Results and models

ScanNet

Backbone	Mem (GB)	Inf time (fps)	AP@0.25 max | mean | ours	AP@0.5 max | mean | ours	Download
MinkResNet34	10.5	8.0	71.5 | 70.7 | 69.7	57.3 | 56.0 | 55.2	model | log

SUN RGB-D

Backbone	Mem (GB)	Inf time (fps)	AP@0.25 max | mean | ours	AP@0.5 max | mean | ours	Download
MinkResNet34	6.3	15.6	64.2 | 63.8 | 64.8	48.9 | 48.2 | 48.2	model | log

S3DIS

Backbone	Mem (GB)	Inf time (fps)	AP@0.25 max | mean | ours	AP@0.25 max | mean | ours	Download
MinkResNet34	23.5	4.2	66.7 | 64.9 | 67.4	45.9 | 43.8 | 45.7	model | log

Citation

@inproceedings{rukhovich2022fcaf3d,
  title={FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection},
  author={Danila Rukhovich, Anna Vorontsova, Anton Konushin},
  booktitle={European conference on computer vision},
  year={2022}
}