From 2D to 3D: Re-thinking Benchmarking of Monocular Depth Prediction

This directory consists of 3D point cloud metrics for evaluating monocular depth prediction proposed in From 2D to 3D: Re-thinking Benchmarking of Monocular Depth Prediction:

• Chamfer distance
• IoU
• F-Score

We provide two implementations: one based on Kaolin library, and another one explicitly implemented using only Pytorch3D kNN function. We also provide the point cloud utilization functions and commonly used 2D and iBims occlusion boundary metrics.

Additional repositories

• Along with this repo, Kaolin library provides nice utilities: https://github.com/NVIDIAGameWorks/kaolin/
• Pytorch3D https://github.com/facebookresearch/pytorch3d
• Occlusion boundary metric adopted from: iBims and https://github.com/MichaelRamamonjisoa/SharpNet
• Other Chamfer Distance implementations:
  • https://github.com/ThibaultGROUEIX/ChamferDistancePytorch
  • https://github.com/chrdiller/pyTorchChamferDistance

References

If you use proposed 3d metrics in monodepth evaluation or wish to refer our arxiv paper, please cite

@article{ornek2022from2dto3d,
  title={From {{2D}} to {{3D}}: Re-thinking Benchmarking of Monocular Depth Prediction},
  author={{\"O}rnek, Evin P{\i}nar and Mudgal, Shristi and Wald, Johanna and Wang, Yida and Navab, Nassir and Tombari, Federico},
  journal={arXiv preprint arXiv:2203.08122},
  year={2022}
}