Table 4 Pixel level labeling accuracies (%) for various algorithms on MSRC-9.
From: Scene Segmentation with Low-Dimensional Semantic Representations and Conditional Random Fields
Method | Object class | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Â | Building | Grass | Tree | Cow | Sky | Aeroplane | Face | Car | Bicycle | Per pixel |
Schroff et al. [49] | 56.7 | 84.8 | 76.4 | 83.8 | 81.1 | 53.8 | 68.5 | 71.4 | 72.0 | 75.2 |
PLSA-MRF [13]a | 74.0 | 88.7 | 64.4 | 77.4 | 95.7 | 92.2 | 88.8 | 81.1 | 78.7 | 82.3 |
CRF[8]a | 73.6 | 91.1 | 82.1 | 73.6 | 95.7 | 78.3 | 89.5 | 84.5 | 81.4 | 84.9 |
LTRF [16] | 78.1 | 92.5 | 85.4 | 86.7 | 94.6 | 77.9 | 83.5 | 74.7 | 88.3 | 86.7 |
RF-CRF [10] | — | — | — | — | — | — | — | — | — | 87.2 |
RLP-CRF [21]b | — | — | — | — | — | — | — | — | — | 88.5 |
Our LRC/CRF-avea | 82.4 | 93.9 | 85.2 | 81.8 | 93.8 | 76.0 | 92.6 | 90.2 | 88.5 | 88.6 |
Our LRC/CRF-min | 79.5 | 90.8 | 87.9 | 77.7 | 90.7 | 72.6 | 91.2 | 82.6 | 95.2 | 86.6 |
Our LRC/CRF-max | 86.6 | 94.7 | 87.7 | 87.7 | 91.8 | 83.5 | 98.8 | 92.4 | 86.3 | 90.7 |