300-W
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
300-W (Common)
0.024NME
180
300-W (Fullset)
2.78Mean Error (%)
174
300-W (Full)
2.87NME
66
300-W Challenging Subset
4.93Mean Error
49
300-W (Full set)
2.83NME
46
300-W public Challenging inter-pupil normalization (test)
6.38NME
46
300-W Common Subset
2.46Mean Error (%)
37
300-W (Challenging set)
3.36NME
32
300-W public Full inter-pupil normalization (test)
4.04NME
29
300-W Public Common subset inter-pupil normalization (test)
3.27NME
28
300-W public (test)
2.62NME
28
300-W (test)
4.94Landmark Prediction Error
12
300-W (Split 1)
2.72NME Common
11
300-W competition 51-points (test)
58.11AUC (0.08 Threshold)
10
300-W Split 2 (test)
2.1NME (Box)
7
300-W competition 68-points (test)
46.56AUC (Threshold 0.08)
7