FS6D-PyTorch icon indicating copy to clipboard operation
FS6D-PyTorch copied to clipboard

Published 20 hours ago • verified publisher icon ethnhe
→

Metadata

FS6D: Few-Shot 6D Pose Estimation of Novel Objects, CVPR 2022

FS6D

This is the official source code for the CVPR 2022 work, FS6D: Few-Shot 6D Pose Estimation of Novel Objects.

Project Page | Arxiv | ShapeNet6D

Raw Source Code & Pre-trained Weights

For those who want the code for reference, the uncleaned raw source code and pre-trained weights can be found here. I will clean it up if I have time.

Introduction & Citation

We study the new open-set few-shot 6D object poses estimation problem: estimating the 6D pose of an unknown object by a few support views without CAD models and extra training. We propose a large-scale synthesis dataset for network pretraining. We also discuss possible solution to the problem and introduce a dense prototypes matching framework. The benchmark for the problem is established to facilitate future research as well.

Please cite FS6D if you use this repository or the ShapeNet6D dataset in your publications:

@InProceedings{he2022fs6d,
  author    = {Yisheng, He and Yao, Wang and Haoqiang, Fan and Qifeng, Chen and Jian, Sun},
  title     = {FS6D: Few-Shot 6D Pose Estimation of Novel Objects},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month     = {June},
  year      = {2022},
}

Installation

Datasets

  • ShapeNet6D: Download the ShapeNet6D dataset from OneDrive.
  • LineMOD: Download the LineMOD dataset from BOP Benchmark.
  • YCB-Video: Download the YCB-Video Dataset from BOP Benchmark.

Training and evaluating

Training on the ShapeNet6D Dataset

Finetuning on the LineMOD Dataset

Finetuning on the YCB-Video Dataset

Evaluating on the YCB-Video Dataset

License

Licensed under the MIT License.

← Metadata

70
Stars
0
Forks
Watchers

Owner

verified publisher icon ethnhe

Metadata

FS6D: Few-Shot 6D Pose Estimation of Novel Objects, CVPR 2022

Back