Bring Doodles to Life: Meta Open-Sources AI Model

Meta has recently open-sourced an artificial intelligence (AI) project designed to transform drawings and doodles into animations. By making the project open-source, Meta aims to encourage developers to contribute to the project. Additionally, this will expand its capabilities and add new features.

Also Read: Meta to Commercialize Generative AI by December

Meta’s Animated Drawing Tool: A Leap Forward in AI

In 2021, Meta introduced a web-based version of its Animated Drawing tool. This enabled users to upload images of their drawn characters and generate short animations. The company uploaded over 6.7 million images during the demo phase. Human reviewers then filtered these images and shared them with Meta’s research team for further analysis.

The Animated Drawing tool employs a combination of object detection models, pose estimation, and image processing-based segmentation techniques to create digital versions of drawings. It then leverages existing graphics methods to animate the objects.

Expanding the Project with Open Source Contributions

Recently, Meta open-sourced the project, providing a dataset of approximately 180,000 drawings and sharing the animation pipeline code. This move will allow more users and developers to contribute to the project, enhancing its capabilities and fostering collaboration.

Animating Children’s Imaginative Drawings

Children’s drawings often feature unique and imaginative characters, requiring a different approach to recognizing the people and objects they depict. While parents or teachers can easily identify the intended subjects in children’s drawings, AI systems often struggle with this task due to the abstract nature of kids’ drawings.

Meta’s researchers have developed an innovative method to use AI for automatically animating children’s hand-drawn figures. By uploading their drawings to the prototype system, children and parents can watch their creations come to life, dancing, skipping, and jumping. Users can even download videos of their animated drawings to share with friends and family.

Overcoming AI Challenges with Children’s Drawings

Many AI tools can handle realistic human images, but children’s drawings introduce a level of variety and unpredictability that complicates identification. Thus, Meta’s project aims to create an AI system that can accurately identify and animate humanlike figures in children’s drawings without human intervention.

To achieve this, the researchers developed a system that teaches AI to recognize humanlike figures in children’s drawings, even when other objects on the page share similar features, colors, or dimensions. The AI then isolates the humanlike figure, ensuring that the animated character closely aligns with the child’s intention. After locating the joints of the figure, the AI enables the animation to move in an appealing way, and the drawing is ready for animation.

Building AI That Understands the World from a Human Perspective

The primary objective of this project is to teach AI to work effectively with children’s drawings. Thus, ultimately bringing the technology closer to understanding the world from a human viewpoint. Meta aims to enhance the capabilities of AI in understanding and animating the creative works of young minds. It is doing this by open-sourcing the project and inviting contributions from developers worldwide,

In conclusion, Meta’s open-source AI project represents a significant advancement in the field of artificial intelligence. Furthermore, AI is also going to play a major role in education, entertainment, and communication. As developers and researchers collaborate on the project, the potential to unlock new ways of interacting with children’s drawings inspires creativity and fosters understanding between humans and AI systems.

Learn More: Decode Meta’s AI Tool For Animating Children’s Drawings