TL;DR
This paper introduces a novel method for automatically converting digital 3D models into "multilayer models" – stacks of 2D images embedded in transparent materials. This technique allows for rapid, cost-effective fabrication of physical 3D objects that offer a natural sense of depth and texture, overcoming limitations of traditional 3D printing for certain object types.
What Problem Does This Paper Address?
Creating physical 3D replicas of digital models can be expensive and time-consuming, especially for complex or delicate objects. Existing methods like 3D printing and multi-axis milling have limitations: they can be costly, slow, and struggle with thin features or disconnected parts. While artists have long created 3D forms by stacking painted glass or acrylic, automating this process for digital models presents significant challenges. Specifically, naively projecting a 3D model onto parallel layers can lead to visible cracks, seams, and artifacts where important surface features are split across layers. Additionally, issues like self-shadowing between layers and light absorption in the fabrication material can degrade the final 3D effect. This paper aims to bridge this gap by providing an automated, high-quality method for generating these multilayer models.
Methods
The core of the paper is an algorithm that converts a 3D surface into a stack of 2D images, each intended for a specific layer in the final physical model. The process is layer-by-layer, from front to back:
- Geometric Warping: Instead of a simple projection, the algorithm warps each layer's content. It computes a "mean visible view direction" for each pixel, considering occlusion from layers above. This direction is then used to sample the original 3D model, effectively stretching the content to avoid cracks and seams within the target viewing cone. A fast approximation using a "flattened" occluding layer and FFT convolution is employed for efficiency.
- Preserving Salient Surface Features: To prevent cutting important features (like eyes or mouths) across layers, a graph-cut approach is used. This allows surface points to be assigned to a layer even if it's not the geometrically closest, prioritizing feature integrity over strict proximity.
- Accounting for Shadowing: To counteract the loss of 3D effect due to shadows cast by opaque pixels onto layers below, the algorithm computes and applies a brightness correction factor based on ambient occlusion. This is approximated using a convolution similar to the geometric warping step.
- Accounting for Absorption: Light absorption and scattering in the transparent medium (glass, acrylic) reduce contrast and brightness in deeper layers. The authors model this using an analytic function and estimate its parameters by measuring color falloff through multiple layers. This allows them to adjust the color gamut of printed layers to compensate for attenuation, ensuring a more uniform appearance.
For volumetric data, a simpler resampling approach is used, leveraging existing volume rendering techniques.
Datasets
The paper does not explicitly mention specific datasets for training or evaluation. Instead, it uses various 3D models (e.g., a bird, a tree, a lion, parrots, a human head) and volumetric datasets (e.g., MRI data) as input for their fabrication process. The results are demonstrated through physical fabrications of these models.
Key Insights and Ideas
- Artistic Inspiration, Engineering Solution: The work draws inspiration from artists who create 3D forms by stacking painted media, but automates this process with sophisticated algorithms.
- "Stretching" Content to Avoid Cracks: The geometric warping technique, by sampling along a "mean visible view direction," is a clever way to make the 2D representation appear continuous and seamless from multiple viewpoints, rather than relying on simple projection which creates artifacts.
- Graph Cuts for Feature Integrity: Using graph cuts to decide layer assignments is a powerful way to ensure that visually important features remain intact on a single layer, even if it means a slight deviation from the closest geometric projection.
- Addressing Volumetric Effects: The explicit consideration of inter-layer shadowing and light absorption is crucial for achieving a convincing 3D illusion in a physically fabricated object, moving beyond purely geometric considerations.
Key Results
The paper demonstrates the successful fabrication of several multilayer models using commercially available printers and materials like glass and acrylic. The resulting models exhibit a natural 3D effect with good depth and texture.
- Speed: The conversion process takes between 5 to 45 seconds per layer, with the majority of time spent on ray/geometry intersections.
- Cost: Multilayer models are presented as significantly cheaper and faster to produce than traditional 3D printing for comparable objects. For example, acrylic tiles cost $2.60 each, and printing takes minutes per layer.
- Capability: The method can fabricate objects with thin features and disconnected parts that are impossible with additive 3D printers or milling machines (e.g., the tree model).
- Visual Quality: The authors show comparisons of their method against simpler projection techniques, highlighting the reduction in cracks and artifacts. They also demonstrate the improvement in visual quality after accounting for shadowing and absorption.
Limitations and Future Work
- Limitations:
- Long, Thin Features: The method can still struggle with very long, thin features that span multiple layers, potentially causing cracks from certain views.
- Hard Directional Lighting: The ambient occlusion-based shadow correction works best with diffuse lighting and breaks down under high-frequency lighting conditions (e.g., a small area light source).
- Surface Properties: The current approach is limited to diffuse surfaces; reproducing directionally-dependent appearance (like specular highlights) is an area for future work.
- Refraction: While the paper mentions that refraction can shift the viewer's perceived position, it largely ignores it for simplicity, assuming the effect is similar to shifting the viewing position.
- Future Work:
- Active Multilayer Displays: Adapting the algorithms for active multilayer displays (e.g., parallel transparent LCDs) to achieve dynamic 3D effects.
- Advanced Inks and Materials: Exploring a wider range of inks and fabrication technologies for more compelling and realistic printouts.
- Complex Surface Shading: Modifying the surface geometry of individual tiles (e.g., by milling) to enable more complex surface shading effects.
- Global Optimization: Investigating global optimization approaches for layer assignment, rather than the current layer-by-layer method, to potentially achieve even better visual results.
Summary updated