Table 5 Pixel level labeling accuracies for various algorithms on Sowerby and Corel.
From: Scene Segmentation with Low-Dimensional Semantic Representations and Conditional Random Fields
Method | Performance | |||||
|---|---|---|---|---|---|---|
Sowerby | Corel | |||||
Accuracy | Training time | Test time | Accuracy | Training time | Test time | |
Shotton et al. [7] | 88.6% | 5 h | 10 s | 74.6% | 12 h | 30 s |
He et al. [2] | 89.5% | Gibbs | Gibbs | 80.0% | Gibbs | Gibbs |
Verbeek and Triggs [8] | 87.4% | 20 min | 5 s | 74.6% | 15 min | 3 s |
Toyoda and Hasegawa [11] | 90.0% | — | — | 83.0% | — | — |
Gould et al. [21]* | 87.5% | — | — | 77.3% | — | — |
Our LRC/CRF* | 89.1% | 8 min | <0.02 s | 77.0% | 7 min | <0.02 s |