Leaderboard

In this leaderboard, you can compare the performance of different algorithms for head pose estimation using our Head Pose Database. You can submit your results and share them with other researchers here.

    We can compare the mean error of the algorithms according to two criteria:
  • Estimation - Ground Truth (Absolute error): The error is obtained as the mean absolute difference between the database ground truth and the submitted estimation.
  • Estimation - Ground Truth Aligned (Differential error): The error is obtained as the mean absolute difference between the database ground truth and the submitted estimation, both of them transformed to be aligned at 0 in the initial frame of each video.

Valid frames: Algorithms may return invalid estimations for some frames (i.e. NaNs). In those cases, estimations are ignored. Percentage of valid estimations is shown.

Sort by:

Rank Name Estimation - Ground Truth Estimation - Ground Truth Aligned Valid
frames (%)
Actions
Roll Yaw Pitch Mean Roll Yaw Pitch Mean

1

Posit + ASM 2D + BFM Model

1.12

2.97

4.04

2.71

0.88

2.50

2.77

2.05

100.00

2

Posit + AAM 2D + BFM Model

1.74

2.30

6.01

3.35

1.04

1.63

2.19

1.62

100.00

3

Posit + ASM 2D + Cylindrical Model

1.14

3.56

5.52

3.40

0.96

2.80

3.03

2.26

100.00

4

Posit + AAM 2D + Cylindrical Model

1.74

3.68

8.83

4.75

1.13

2.76

2.89

2.26

100.00