The following metrics are consistently used in our benchmark:
-
Mean Corruption Error (mCE):
- The Corruption Error (CE) for model
$A$ under corruption type$i$ across 3 severity levels is:$\text{CE}_i^{\text{Model}A} = \frac{\sum((1 - \text{mIoU})^{\text{Model}A})}{\sum((1 - \text{mIoU})^{\text{Baseline}})}$ . - The average CE for model
$A$ on all$N$ corruption types, i.e., mCE, is calculated as:$\text{mCE} = \frac{1}{N}\sum\text{CE}_i$ .
- The Corruption Error (CE) for model
-
Mean Resilience Rate (mRR):
- The Resilience Rate (RR) for model
$A$ under corruption type$i$ across 3 severity levels is:$\text{RR}_i^{\text{Model}A} = \frac{\sum(\text{mIoU}^{\text{Model}A})}{3\times (\text{clean-mIoU}^{\text{Model}A})} .$ - The average RR for model
$A$ on all$N$ corruption types, i.e., mRR, is calculated as:$\text{mRR} = \frac{1}{N}\sum\text{RR}_i$ .
- The Resilience Rate (RR) for model
Corruption | Light | Moderate | Heavy | Average | ||
---|---|---|---|---|---|---|
Fog | 54.57 | 49.31 | 32.67 | 45.52 | 123.45 | 68.93 |
Wet Ground | 62.55 | 59.56 | 53.54 | 58.55 | 90.09 | 88.66 |
Snow | 49.60 | 49.41 | 48.88 | 49.30 | 108.52 | 74.65 |
Motion Blur | 37.61 | 32.75 | 28.69 | 33.02 | 99.85 | 50.00 |
Beam Missing | 63.76 | 59.46 | 54.63 | 59.28 | 93.22 | 89.76 |
Crosstalk | 26.66 | 22.02 | 18.76 | 22.48 | 186.08 | 34.04 |
Incomplete Echo | 62.55 | 59.56 | 53.54 | 58.55 | 90.96 | 88.66 |
Cross-Sensor | 63.34 | 59.77 | 40.74 | 54.62 | 84.11 | 82.71 |
-
Summary:
$\text{mIoU}_{\text{clean}} =$ 66.04%,$\text{mCE} =$ 109.54%,$\text{mRR} =$ 72.18%.
@article{puy2023waffleiron,
title = {Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation},
author = {Puy, Gilles and Boulch, Alexandre and Marlet, Renaud},
journal = {arXiv preprint arxiv:2301.10100}
year = {2023}
}