Skip to content
This repository has been archived by the owner on Nov 1, 2024. It is now read-only.

Latest commit

 

History

History
71 lines (50 loc) · 3.64 KB

RESULTS.md

File metadata and controls

71 lines (50 loc) · 3.64 KB
Raw

Results for released models

Note that this repository is a large refactoring of the original code to allow for public release and to integrate with pytorch3d. Hence the models/datasets are not necessarily the same as that in the paper, as we cannot release the saved test images we used. To compare results, we recommend comparing against the numbers and models in this repo for fair comparison and reproducibility.

These models have been trained with the same learning rate (0.0001) and number of epochs. If you want to train the baselines differently (for example in the paper we found that the voxel based methods were highly sensitive to learning rates, so used a model trained with a lower learning rate), look at the options in ./../submit.sh.

You can use these numbers to:

  1. Compare against your models
  2. Verify your setup is indeed correct

Results on RealEstate

PSNR SSIM Perc SIM
SynSin 22.31 0.74 1.18
SynSin+ 22.83 0.75 1.13
ViewAppearance [1] 17.05 0.56 2.19
Tatarchenko [2] 11.35 0.33 3.95
StereoMag [4] 25.34 0.82 1.19
3DPaper [5] 21.88 0.66 1.52

Results on Matterport

PSNR SSIM Perc SIM
SynSin 20.91 0.72 1.68
ViewAppearance [1] 15.87 0.53 2.99
Tatarchenko [2] 14.79 0.57 3.73

Results on Replica

PSNR SSIM Perc SIM
SynSin 21.94 0.81 1.55
ViewAppearance [1] 17.42 0.66 2.29
Tatarchenko [2] 14.36 0.68 3.36

Results on KITTI

PSNR SSIM Perc SIM
SynSin* 16.70 0.52 2.07
SynSin+* 16.73 0.52 2.05
ViewAppearance [1] 14.21 0.43 2.51
Tatarchenko [2] 10.31 0.30 3.48
ContView [3] 16.90 0.54 2.21

References

The implemented models are based on:

*: Using inverse depth as opposed to a uniform sampling. This is better if there is a long tail distribution of the true depths (as in the KITTI case).

+: Leaving the model to run for longer than the paper for a small boost in results.

[1] Zhou, Tinghui, et al. "View synthesis by appearance flow." ECCV, 2016.

[2] Dosovitskiy, Alexey, et al. "Learning to generate chairs with convolutional neural networks." CVPR, 2015.

[3] Chen, Xu, et al. "Monocular Neural Image Based Rendering with Continuous View Control." ICCV, 2019.

[4] Zhou, Tinghui, et al. "Stereo Magnification: Learning View Synthesis using Multiplane Images." SIGGRAPH, 2018.

[5] Code based on work by folks at Facebook. The code used is an early version of the 3D Photos work.