New: WildDash 2 with 4256 public frames, new labels & panoptic GT!
See also: RailSem19 dataset for rail scene understanding.
For all metrics, higher scores are better. To participate in the benchmark, check our submission instructions.
| Meta AVG | Classic | Negative | Impact (IoU class) | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Algorithm | IoU Class | IoU Class | iIoU Class | IoU Cat. | iIoU Cat. | IoU Class | Blur | Coverage | Distortion | Hood | Occ. | Overexp. | Particles | Screen | Underexp. | Var. |
| SN_DN161_fat_pyrx8 | 46.8% | 51.0% | 43.9% | 71.4% | 65.5% | 32.6% | -7% | -11% | -5% | -9% | -3% | -2% | -7% | -22% | -8% | -8% |
| SN_DN161s3pyrx8 | 45.6% | 49.8% | 41.6% | 71.3% | 65.3% | 31.0% | -10% | -6% | -6% | -10% | -3% | -3% | -6% | -20% | -9% | -10% |
| SN_RN152pyrx8_RVC | 45.4% | 48.9% | 42.7% | 70.1% | 64.8% | 32.5% | -6% | -7% | -5% | -7% | -1% | -2% | -7% | -19% | -11% | -3% |
| seamseg_rvcsubset | 37.9% | 41.2% | 37.2% | 63.1% | 58.1% | 30.5% | -16% | -17% | 0% | -7% | -4% | -14% | -18% | -31% | -14% | -7% |
| Tong | 37.2% | 41.0% | 41.2% | 65.2% | 53.5% | 26.0% | -18% | -9% | -5% | -16% | -2% | -13% | -12% | -24% | -10% | -1% |
| seamseg_mvd_ss | 37.1% | 41.3% | 36.9% | 63.4% | 55.7% | 26.6% | -15% | -14% | 0% | -11% | -4% | -11% | -30% | -36% | -20% | -10% |
| SIW | 36.5% | 41.0% | 38.6% | 65.8% | 53.1% | 24.1% | -16% | -17% | -6% | -14% | -2% | -7% | -19% | -23% | -10% | -6% |
| hs1 | 35.7% | 40.0% | 38.0% | 64.8% | 52.3% | 23.0% | -17% | -10% | -8% | -18% | -1% | -15% | -11% | -27% | -9% | -9% |
| MSeg1080_RVC | 35.2% | 38.7% | 35.4% | 65.1% | 50.7% | 24.7% | -15% | -11% | -9% | -19% | -3% | -14% | -6% | -25% | -8% | -13% |
| tong_test | 34.6% | 38.7% | 36.3% | 63.6% | 50.8% | 22.5% | -17% | -7% | -8% | -15% | -1% | -12% | -9% | -24% | -11% | -20% |
| hs | 34.4% | 38.4% | 36.2% | 64.2% | 52.1% | 22.3% | -19% | -11% | -8% | -18% | 0% | -13% | -15% | -29% | -11% | -6% |
| EffPS_b1bs4sem_RVC | 32.2% | 35.7% | 24.4% | 63.8% | 56.0% | 20.4% | -10% | -6% | -4% | -7% | -1% | -7% | -10% | -25% | -8% | -6% |
Methodology:
Our benchmark evaluates the negative Impact of common visual hazards on algorithm output performance. It is calculated by this formula:
impact = min(metriclow,metrichigh) / max(metricnone,metriclow) - 1.0
The metricsnone/low/high are evaluated on subsets of the benchmark dataset that correspond to the identified severity of the hazard (e.g. the subset Blurhigh contains images which have a lot of blur visible). Positive impacts are truncated to zero.
An impact of -10% at Blur translates to an expected performance degradation for the algorithm of 10 percent when there is a considerable blur in the input image as opposed to supplying the same algorithm a similar image without noticeable image blur.
These are all currently evaluated hazards:
Blur: Image is noticeably affected by blur (e.g. motion blur, defocusing, compression artifacts...)
Coverage: Normally visible parts of the road are covered (e.g. unusual lane markings, snow, leaves...)
Distortion: Visible lens distortion
Hood: Ego-vehicle is visible, non-windscreen parts (e.g. car hood, mirrors)
Occl: Objects are partially occluded or cut off by image border
Overexp.: The scene is overexposed
Particle: Particles in the air obstruct the view (e.g. heavy rain, snow, fog)
Screen: The windscreen is interfering (e.g. interior reflections, wipers, rain on the windscreen,...)
Underexp.: The image is underexposed
Variation: Intra-class variations within the image (i.e. unusual representations of labels like unique cars)
More details on evaluation metrics and negative test cases can also be found on the FAQ page.
