Animating Childlike Drawings with 2.5D Character Rigs

TL;DR

This work addresses the challenge of animating childlike drawings in 3D while preserving their distinctive style. It introduces a view-dependent 2.5D character model that starts from a single drawing and annotations, then retargets arbitrary 3D skeletal motion onto the figure in real time, using left/right views and keyview-transforms. A key contribution is the projection-plane optimization that mitigates flailing and dampening, along with an annotation-driven rigging workflow validated through perceptual studies and ablations. The approach enables rapid, accessible animation for mixed-reality and 3D contexts, offering a viable stylistic alternative to full 3D models for many character scenarios. Limitations include bidpedal constraints and motion types with strong 3D depth requirements, suggesting future work in broader motion support and AI-assisted annotation.

Abstract

Drawing is a fun and intuitive way to create a character, accessible even to small children. However, animating 2D figure drawings is a much more challenging task, requiring specialized tools and skills. Bringing 2D figures to 3D so they can be animated and consumed in immersive media poses an even greater challenge. Moreover, it is desirable to preserve the unique style and identity of the figure when it is being animated and viewed from different perspectives. In this work, we present an approachable and easy-to-create 2.5D character model and retargeting technique that can apply complex 3D skeletal motion, including rotation within the transverse plane, onto a single childlike figure drawing in a style-preserving manner in realtime. Because our solution is view-dependent, the resulting character is well-suited for animation in both 2D and 3D contexts. We also present a novel annotation study motivating our system design decisions and a pair of user studies validating the usefulness and appeal of our solution. We showcase the generality of our system in a range of 2D and 3D applications.

Figures (9)

• Figure 1: Frequency with which annotators references specific body parts and specified part traits when determining figure orientation.
• Figure 2: Example of the different figure annotations and part-level attributes necessary to create the 2.5D character rigs.
• Figure 3: The part regions, masks, and textures comprising the model's left view. Each part region can be thought of as existing in a different layer (color coded here). The right view is similarly constituted, but left-facing parts, such as hair and nose, are horizontally flipped.
• Figure 4: View unit circle. Root view vector (define as ground projection of vector from camera to character) is always at 0.Change skeleton to pose. When view angle = 0, character is looking away. pi/2 = looking left. pi = looking directly at character 3pi/2 = looking right.
• Figure 5: Examples of the different part traits mentioned by annotators. The extension of the tail and nose in a suggest (conflicting) orientations; in b the position of the eyes within head and pupils within eye suggest it; in c the limbs are drawn as though viewing a character facing right; d-f show less frequently mentioned cues of limb attachment point, perspective-based scaling, and occlusion, respectively.