https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136990277.pdf

单看指标的话,相比他们无监督那版结果,整体iou我们是高的,高大概10个点。

他们自己合成的数据,没有公开数据和code,要一下数据。

compare with Multi-Echo Denoising in Adverse Weather, 23

Untitled

Untitled

θ(·) and ϕ(·) denote PR-Net and RD-Net.

PR-Net: predicate

I is a binary mask that only selects loss from the blanked points

Untitled

Untitled

Untitled

References

  1. SLiDE: Self-supervised LiDAR De-snowing through Reconstruction Difficulty, eccv 22