Video support
Merge request reports
Activity
mentioned in merge request bob.bio.base!264 (merged)
Finally, this is in a good state for merge. The vanilla-biometrics pipeline was not ready to handle
VideoLikeContainer
. Follow below a set of updates I've done along with the stack.-
bob.bio.base
Updated theenrolment
andscoring function
to handle a list of features coming from video sequences, which are naturally 3d sequences. -
bob.bio.base
Updated thescore_reduction_operation
to properlly reduce the scores that comes from a video sequence -
bob.bio.face
. Some features extractors were returningzXnXd
numpy arrays oncememory_demanding==True
. This is a bug. Feature extractors (the ones where features are numpy arrays) should always returnnXd
no matter what is set inmemory_demanding
-
bob.bio.video
. Created a wrapper that given a scikit-learn pipeline, wraps all it's estimators withVideoWrapper
. Also, the entry-pointvideo-wrapper
was created. In this way, we can automatically use all our face recognition baselines in video databases. -
bob.bio.video
. Ported the YouTube faces dataset to the new Database API. In this way,bob.db.youtube
can rest in peace. -
bob.bio.video
. Documented thebob.bio.video
package containing instructions on how to use the video support and a FAQ section describing how to customize the frame selector.
I'm running some experiments of
fold0
with some of our baselines and seems that the plots are sensible. eval_plots.pdfThis depends on:
-
This closes #20 (closed) #19 (closed) #15 (closed)
Last update with our baselines on
fold0
and the default cropping and frame selection setup.mentioned in merge request bob.bio.face!133 (merged)
mentioned in commit f90a96e0
mentioned in issue #20 (closed)
mentioned in issue #19 (closed)
mentioned in issue #15 (closed)
Hey @mguenther,
@amohammadi is working on the conda setup this week, so we currently don't have beta packages to be installed. You would need to
buildout
themmentioned in issue bob#269 (closed)