Could the implementation of the age/gender model inference be wrong?

[lunzueta]

There is a brief explanation here about how the model was trained that says: "The network is able to recognize age of people in [18, 75] years old range, it is not applicable for children since their faces were not in the training set."

Then, in the code the operation done to get that number is:

ageBlob->buffer().as<float*>()[idx] * 100

I assume that output of ageBlob is a number in the range [0, 1], and if it is then multiplied by 100 it could give an age in the range [0, 100]. Taking into account the description given above, shouldn't it be something like:

18 + (75 - 18) * ageBlob->buffer().as<float*>()[idx]

Could we know more details about the training of this model?

Thanks

[dkurt]

"it is not applicable" is not "doesn't work". See outputs section:

1. name: "age_conv3", shape: [1, 1, 1, 1] - Estimated age divided by 100.

That means that network has never seen the examples of age < 18. In different words, if your model has been trained to classify cats and dogs it can predict that a fox is a dog with confidence 0.3. That means that model can recognize an animal which looks similar to ones it saw before. But the models is not applicable for foxes.

[lunzueta]

OK, thank you for the clarifications.