generating .npy files about sigver_wiwd

[luizgh]

Hello Tahir. In this repository we released the code / trained models for extracting features for signatures, and not the code for performing classification. In the example.py, the comparison to the .npy files is just to make sure that the results you obtain is the same as the results I obtained. It is not classifying the signature of the user.

For actual classification, the simplest approach is computing the distance (in feature space) between a query signature and a reference. This is easy to implement (see the jupyter notebook example), but not very powerful. Most commonly, people train "Writer-dependent classifiers" (one binary classifier for each user). You may find it useful to read more about this topic in these papers:

https://arxiv.org/abs/1507.07909
https://arxiv.org/abs/1705.05787

[TahirJanjua]

Thank you so much for your guidance . I shall be following the links you send and implementing what you suggested .I have found your repository very helpful.  Best RegardsTahir  Sent from Yahoo Mail on Android On Tue, 8 Jan 2019 at 11:39 pm, Luiz Gustavo Hafemann<[email protected]> wrote: Hello Tahir. In this repository we released the code / trained models for extracting features for signatures, and not the code for performing classification. In the example.py, the comparison to the .npy files is just to make sure that the results you obtain is the same as the results I obtained. It is not classifying the signature of the user. For actual classification, the simplest approach is computing the distance (in feature space) between a query signature and a reference. This is easy to implement (see the jupyter notebook example), but not very powerful. Most commonly, people train "Writer-dependent classifiers" (one binary classifier for each user). You may find it useful to read more about this topic in these papers: https://arxiv.org/abs/1507.07909 https://arxiv.org/abs/1705.05787 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.