module art.videodl

Short summary

module code_beatrix.art.videodl

Fonctions proposant de traiter des vidéos avec des traitements compliqués type deep learning.

source on GitHub

Functions

function	truncated documentation
video_map_images	Applies one complex process to a video such as extracting characters from videos and removing the backaground. It …
video_map_images_detect	Blurs people faces. Uses function detectmultiscale. …
video_map_images_people	Extracts characters from a movie. The movie is composed with an image and a mask. …

Documentation

Fonctions proposant de traiter des vidéos avec des traitements compliqués type deep learning.

source on GitHub

code_beatrix.art.videodl.video_map_images(video_or_file, name, fLOG=None, **kwargs)

Applies one complex process to a video such as extracting characters from videos and removing the backaground. It is done image by image. Applique un traitement compliqué sur une séquence d’images telle que la séparation des personnages et du fond.

Paramètres:

• video_or_file – string or VideoClip

• name – name of the processing to do, see the list below

• fLOG – logging function

• kwargs – additional parameters

Renvoie:

VideoClip

List of available treatments:

• 'people': extracts characters from a movie. The movie is composed with an image and a mask. Parameters: see @fn video_map_images_people.

• 'detect' : blurs or put a rectangle around faces, uses opencv

Avertissement

A couple of errors timeout, out of memory… The following processes might be quite time consuming or memory consuming. If it is the case, you should think of reducing the resolution, the number of frames per seconds (fps). You can also split the video and process each piece independently and finally concatenate them.

source on GitHub

code_beatrix.art.videodl.video_map_images_detect(video_or_file, fps=None, with_times=False, logger=None, dtype=None, scaleFactor=1.3, minNeighbors=4, minSize=(30, 30), action='blur', color=(255, 255, 0), haar=None, fLOG=None)

Blurs people faces. Uses function detectmultiscale. Relies on opencv. Floute les visages.

Paramètres:

• video_or_file – string or VideoClip

• fps – see video_enumerate_frames, faces are detected in each frame returned by video_enumerate_frames

• with_times – see video_enumerate_frames

• logger – see video_enumerate_frames

• dtype – see video_enumerate_frames

• fLOG – logging function

• scaleFactor –

  see detectmultiscale

• minNeighbors –

  see detectmultiscale

• minSize –

  see detectmultiscale

• haar – shape classifier to load, face by default, see below

• action – to blur, to put a rectangle around the detected zone… see below

• color – rectangle color if action is 'rect'

Renvoie:

VideoClip

Only haarcascade_frontalface_alt.xml is provided but you can get more at haarcascades.

Parameter action can be:

• 'blur': to blur faces (or detector zones)

• 'rect': to draw a rectangle around faces (or detector zones)

Faces in a yellow box in a video

The following example uses opencv to detect faces on each image of a video and put a yellow box around each of them.

from code_beatrix.art.videodl import video_map_images
from code_beatrix.art.video import video_save, video_extract_video

vide = video_extract_video(vid, 0, 5 if __name__ == "__main__" else 1)
vid2 = video_map_images(
    vide, fps=10, name='detect', action='rect',
    logger='bar', fLOG=fLOG)
exp = os.path.join(temp, "people.mp4")
video_save(vid2, exp, fps=10)

The following video is taken from Charlie Chaplin’s movies.

unable to find 'face.mp4'

source on GitHub

code_beatrix.art.videodl.video_map_images_people(video_or_file, resize=('max2', 400), fps=None, with_times=False, logger=None, dtype=None, class_to_keep=15, fLOG=None, **kwargs)

Extracts characters from a movie. The movie is composed with an image and a mask. Extrait les personnages d’un film, le résultat est composé d’une image et d’un masque transparent qui laissera apparaître l’image d’en dessous si cette vidéo est aposée sur une autre.

Paramètres:

• video_or_file – string or VideoClip

• resize – see predict

• fps – see video_enumerate_frames

• with_times – see video_enumerate_frames

• logger – see video_enumerate_frames

• dtype – see video_enumerate_frames

• class_to_keep – class to keep from the image, it can a number (15 for the background, a list of classes, a function which takes an image and the prediction and returns an image)

• fLOG – logging function

• kwargs – see DLImageSegmentation

Renvoie:

VideoClip

Avertissement

A couple of errors timeout, out of memory… The following processes might be quite time consuming or memory consuming. If it is the case, you should think of reducing the resolution, the number of frames per seconds (fps). You can also split the video and process each piece independently and finally concatenate them.

Extract characters from a video.

The following example shows how to extract a movie with people and without the background. It works better if the contrast between the characters and the background is high.

from code_beatrix.art.video import video_extract_video, video_save
from code_beatrix.art.videodl import video_map_images

vide = video_extract_video("something.mp4", 0, 5)
vid2 = video_map_images(vide, fps=10, name="people", logger='bar')
video_save(vid2, "people.mp4")

The function returns something like the the following. The character is wearing black and the background is quite dark too. That explains that the kind of large halo around the character.

unable to find 'videodl.mp4'

source on GitHub