Generate a synthetic database using artificial neural networks

Offering extra earnings. For research purposes, it is necessary to generate a synthetic database of video images of people's faces. Images of faces must be placed in a short video lasting 10 seconds.

As the original video, you can take any cuts from YouTube. As the person who will superimpose on the original image, you can take faces from any open databases.

Outcome of work:
1) a video database with synthetic faces instead of the original ones with a size of at least 10,000. Each video is 10 seconds long.
(2) original video database
(3) base generation source code
(4) references to all databases used

Acceptance criteria:
(1) Face resolution in the image: on average, video is at least 100 pixels of eye spacing (recommended 150)
(2) the sharpness of the generated image shall be comparable to the original (visually)
(3) the overlapping face should not be visible visually on most of the video (small distortions of the superimposed image are allowed when the head is turned more than 15 degrees to the side)

The cost and timing are contractual. In case of obtaining the quality of the generated image above the specified in the Criteria, additional payment is possible at the discretion of the Customer.