Overall, SAFR usually produces a mean absolute age error of +/- 3.88 years. This is an aggregate estimate and results will vary with each individual and may not fall within these ranges.
The break down of expected aggregate age detection errors by age group is as follows:
- 0-1 ±1 year
- 1-3 ±0.4 years
- 3-11 ±1.46 years
- 11-20 ±2.12 years
- 20-30 ±1.44 years
- 30-45 ±1.73 years
- 45-65 ±1.6 years
- 65-150 ±9.44 years
For comparison, in published scientific research, the highest performance we were able to find is a mean absolute error of 3.25 years.
Age estimation confidence is not available. This is due to the nature of the model. For best accuracy, input images should be of high quality with good sharpness, contrast and center pose and low/no occlusion.