Similarity calculation using different metrics

[ahshamsi]

Current approach to calculate similarity is converted to liner result .. which may return high matching score for false cases.

Would it be possible to add options from admin to return similarity result using different metrics such as Cosine, Euclidean Distance and L2 form.

[pospielov]

We use Euclidean Distance to calculate similarity. We chose it because facial recognition libraries use it; because of this, we can guarantee that your accuracy is the same as those libraries. It's not easy to add more metrics now, we will consider adding more metrics in future.