face-morphing-multiple-images

Credit to: https://github.com/Azmarie/Face-Morphing

This repo is an upgrade where we can morph multiple images into video.

Check out Azmarie's repo first to understand all the details.

Execution

1. Place your images in a folder, for example raw_images
2. Use the command python code/utils/align_images.py raw_images/ aligned_images/ --output_size=1024 to select images from "raw_images" with a face and align/rotate them and put them in "aligned_images".
3. Launch the command to create the video :

python code/__init__.py --folder aligned_images --output video_output.mp4 --duration 4

Note that this will create temporary videos (--tmpfolder) and then combine them into one video (--output).

Results :

Details to call code/init.py

• --img1 : The First Image (not necessary when --folder is used)
• --img2 : The Second Image (not necessary when --folder is used)
• --folder : The folder with all images to morph (not necessary when --img1 and --img2 are used)
• --duration : The duration of morphing from one image to the other.
• --frame : The frame rate of the encoding.
• --output : Final video path.
• --tmpfolder : Folder to store intermediate videos.

How it works on multiple videos?

1. The program goes through all the images in --folder with a for loop
2. For each 2 images we apply what was done by Azmarie's original repo and this outputs a video. We store them in --tmpfolder
3. We append to a text file named imageslist.txt the names of the videos like so : file '<filename_of_the_video>'
4. After dealing with all images we can encode one big video using imageslist.txt and the right ffmpeg command:

ffmpeg -f concat -safe 0 -i imageslist.txt -c copy output.mp4

Note: This is not the most optimized way to do it but it works. It would be better to directly encode one video but it would be a lot of work on Azmarie's work to do.