Sketch based search in DigiKam

[Carl Schwan]

Dear DigiKam developers,

I'm a student in the university of Saarbrücken (Germany) (and a KDE
contributor ;) ) and for a university project, I and 2 other students will
need to implement a deep learning application and as subject we choose a
sketch base image retrival.

We have two months to finish, and because the goal of this course isn't to
create a fancy GUI to draw the sketch and an advanced database and saw that
DigiKam already has this feature, we will probably fork the DigiKam GUI and
replace the current implementation of the sketch based search in DigiKam with
a new one.

Do the DigiKam developers, think it could be easy to replace the curent sketch
based implementation with another writen in python and pytorch?

And are the DigiKam developers interested in merging the new implementation to
DigiKam, if we archive better result than the current implementation?

We didn't start our implementation yet, but we already have a project proposal
(in the attachment), where we looked at multiple research papers.

Best Regards,

Carl Schwan

[Gilles Caulier-4]

Le mar. 4 juin 2019 à 08:15, Carl Schwan <[hidden email]> a écrit :

Dear DigiKam developers,

Hi,

 

I'm a student in the university of Saarbrücken (Germany) (and a KDE
contributor ;) ) and for a university project, I and 2 other students will
need to implement a deep learning application and as subject we choose a
sketch base image retrival.

We have two months to finish, and because the goal of this course isn't to
create a fancy GUI to draw the sketch and an advanced database and saw that
DigiKam already has this feature, we will probably fork the DigiKam GUI and
replace the current implementation of the sketch based search in DigiKam with
a new one.

Do the DigiKam developers, think it could be easy to replace the curent sketch
based implementation with another writen in python and pytorch?

Definitively no. whole digiKam is written in C++ for performance reasons and to still homogeneous with the rest of the implementation (1.2 M lines of code).

 

And are the DigiKam developers interested in merging the new implementation to
DigiKam, if we archive better result than the current implementation?

If the new one is written in C++ with Qt, we can take a look. Of course the new one must improve the older one.

 

We didn't start our implementation yet, but we already have a project proposal
(in the attachment), where we looked at multiple research papers.

Best

Gilles Caulier