--- Comment #173 from ---

I would to know if the Google Summer of Code 2019 branch about face recognition
port to OpenCV DNN will solve this issue.

Typically, i currently perform a bug triage in Bugzilla, especially about
duplicates files. I discovered that more than 150 files are open about Face
management crashes. It's enormous.

Here i test in-deed the new face detection and recognition code, fully

Even if recognition need more adjustment to improve results, the face detection
is fabulous. Compared to previous one, the results are more than expected. The
neuronal network with a previous training for faces performed by OpenCV team is
just wonderful. It work in all cases, including the critical shot conditions.

I tested with plenty of different collections, all  the detection is very
stable. It never crash with or without multi-core support. 

To test you need :

- to checkout the code from development branch :

git clone digikam
cd digikam
git checkout -b gsoc19-face-recognition remotes/origin/gsoc19-face-recognition
cd build 
make -j4
make install/fast

Warning this will overwrite your current digiKam build from git/master branch.
There is no risk to re-install master later to go back. only internal source
code is touched in this developement branch. There is no difference with
installed files compared with git/master.

- i recommend to run this digiKam version in a new account. This will create a
fresh database. Import your collection which introduce crash during faces scan.

- Go to People tab from left sidebar. Select Scan for faces only. Scan whole
collection (no need to scan tags). In advanced options, don't forget to select
DNN module.

- Start face scan and wait to see if the operation can be completed without
crash. Look also the face detection results to see if you have not too much
false positives.


Gilles Caulier

You are receiving this mail because:
You are watching all bug changes.

Reply via email to