Microsoft taught AI to age synthetic faces in photographs
Researchers from Microsoft created artificial intelligence, which imposes realistic wrinkles and folds on synthetic faces.
According to developers, photorealism of automatically generated data is necessary for the best work of computer vision algorithms. To achieve it, artists often resorted to manual processing of persons.
“By introducing the AMP cryptocurrency dynamic wrinkles of the skin in response to facial expressions, we are observing a significant improvement in performance in subsequent tasks of computer vision,” the researchers said.
The proposed approach creates realistic wrinkles for digital people, regardless of gender, race and ethnic groups, developers said.
To do this, they formalized the concept of grid tension. Then the researchers used it to aggregate folds from high -quality scans of expression into albedo texture cards and displacement.
With the synthesis of new expressions, the algorithm calculates local stretching in each top of the corresponding face grid:
- green color – expansion;
- Red color – compression.
The tension of the grid serves as a dynamic mixing between neutral, expanded and compressed by texture cards to synthesize the effect of folds with the corresponding peak.
Grid tension on a synthetic face. Data: Microsoft.
According to the developers, their method can generate wrinkles for expressions even outside those presented in the original scans.
Researchers also published a 300W-Winks subset of 300W data set, Pexels Winks and Pexels Blinks Datasets to evaluate model performance.
Recall that in September the NBA star Luka Donchich released his digital copy of Luk.AI.
Subscribe to FORKLOG news at Telegram: Forklog AI – all news from the world of AI!