Assignment 1 - Paper Reading #5

Intro:

• Paper reading #5: 2.5D Cartoon Models

Reference information:

• Alec Rivers, Takeo Igarashi, and Frédo Durand. 2010. 2.5D cartoon models. ACM Trans. Graph. 29, 4, Article 59 (July 2010), 7 pages. DOI=10.1145/1778765.1778796 http://doi.acm.org.lib-ezproxy.tamu.edu:2048/10.1145/1778765.1778796

Author Bios: 

• Alec Rivers (middle)

• Since 2008, has been a PhD student in the Computer Graphics Group at MIT
• Undergraduate at Cornell University
• Summer of 2007, worked at the University of Tokyo in the Igarashi User Interface Lab
• Spring of 2008, worked as a computer science teacher at Stuyvesant High School in New York City
• Summer of 2009, worked at Tsinghua University in the CG&CAD laboratory 

• Takeo Igarashi

• Professor at department of computer science, the University of Tokyo
• Directing JST ERATO Igarashi Design Interface Project
• Received Ph.D from department of information engineering, the University of Tokyo in March of 2000. 
• Research interest is in user interfaces in general and interaction techniques for 3D graphics. 
• Known as the inventor of Teddy (sketch-based 3D modeling) in graphics
• Received ACM SIGGRAPH 2006 significant new researcher award.

•  Frédo Durand

• Associate professor in Electrical Engineering and Computer Science at MIT
  
  • Member of the Computer Science and Artificial Intelligence Laboratory
  • Received his PhD from Grenoble University, France, in 1999
  • From 1999 till 2002, he was a post-doc in the MIT Computer Graphics Group

Summary:

This paper studies a way to bring 2D cartoon objects and characters into 3D, by giving them the ability to be rotated and be viewed from any angle. It shows how a 2D vector art drawing of a cartoon from different key views (like the first 3 pictures below) can be used to generate a 2.5D cartoon model (4th picture below), which can be used to simulate 3D rotations and generate drawings of the 2D cartoon from any view (5th picture below) using shape interpolation algorithms. Since 2.5D cartoon models are easier to create than a full 3D model, this paper gives insight on how this type of 2.5D modelling can be used to not only render objects in real time, due to the decreased amount of calculations, but how it can be used in non-accelerated 3D devices and applications like cell phones and flash.

Related work not referenced in the paper:

1. Real-time pen-and-ink illustration of landscapes - This paper presents a non-photorealistic rendering pipeline that supports pen-and-ink drawings for, but not limited to, complex landscape scenes in real time. It describes clustering, which enables novel approaches to efficient/coherent rendering of stylized silhouettes, hatching, and abstract shading. Relevant to current work because it studies different techniques in graphics to promote real-time rendering of images.
2. Video tooning - This paper describes a technique for transforming an input video into a highly abstracted, spatio-temporally coherent cartoon animation with many different styles. It is related to the current paper because it talks about a technique for animating objects into cartoon.
3. New dimension for sketches - This paper proposes a method for creating "smart" sketches of free form 3D objects by using a 2D sketch of the object drawn by the user with information about its 3D structure.This paper relates to the current one by studying how to transform 2D drawing into 3D drawings.

4. Real-time watercolor illustrations of plants using a blurred depth test - This work presents techniques to create high-quality 2D watercolor illustrations of plants and focuses on the real-time rendering. This work is relevant because it talks about transforming a 3D object into a 2D object, opposite of current paper.

   
   

   5. A stylized cartoon hair renderer - This paper describes a new hair rendering technique for anime characters with a goal of improving current cell shaders by using a new hair model and shader. The hair renderer is based on a painterly rendering algorithm (uses a large amount of particles) and the hair model is rendered twice: (1) for generating the silhouettes and (2) for shading the hair strands. Most rendering steps are performed on the GPU to take advantage of recent graphics hardware. This paper is relevant because it is using a technique to render images more efficiently.

   
   

   6. GPU-based rendering and animation for Chinese painting cartoon - This paper presents a real-time rendering system for drawing Chinese ink-and-wash cartoons. The rendering process takes advantage of the GPU, including interior shading, silhouette extracting, and background shading. Talks about taking advantage of accelerated hardware and real-time rendering of images.

   
   

   7. Programmable motion effects - Talks about depicting motion in the rendering process by sampling the position of an object and connecting edges in two adjacent samples in bilinear patches. Relevant because of talk about the rendering process and improving animation.

   
   8. Performance-driven hand-drawn animation - This work talks about sketching "hand-drawn" animations in real time using algorithms that perform multi-way morphs to generate real-time animations that mimic the expressions of a user. Similar in the sense of trying to render animations in
   

   
   9. Hardware-accelerated real-time rendering for 3D Sumi-e painting - This paper presents a method for rendering a 3D Sumi-e image in real-time using regular graphics hardware. This is very similar to the current paper because it studies how to render 3D images in real-time and its past research used 2D and 2.5D techniques.
   

   
   10. Cartoon rendering of smoke animations - This paper describes a way to generate cartoon style animations of smoke by using the output of  physics-based simulator and rendering based on the differences in particle depth. Relevant because of talk about rendering of objects and rendering techniques.
   

All of these works are related to the current paper because they all talk about a way to perform graphical animation, maximizing the use of the GPU, and trying to perform rendering of objects, both 2D and 3D, in real-time.

Evaluation:

The study was evaluated quantitatively and qualitatively using subjective (people may have different opinions on whether the rendering is plausible) measures. A table was used to analyze the amount of strokes, average key views per stroke (views from which stroke can be seen by camera), and the number of overlap constraints (given an order in the z direction to determine depth) used to render a 2.5 drawing of each object. The table is shown below.

Model | # Strokes | Avg. # Key Views per Stroke | # Overlap Constraints
Professor | 28 | 3.8 | 17
Face | 18 | 3.6 | 18
Alien | 32 | 3.125 | 9
Dog | 14 | 3.2 | 5

A rendering of 4 2D objects into 2.5 drawings was done to measure the study qualitatively. All of the objects were able to be rendered in any view given the 3 key views. This shows that only 3 key views are needed to generate plausible renderings of a 3D object, from a 2.5D interpolated object. The results also showed that significantly less and simple calculations are needed to render the 2.5 drawing, so it can be done in real-time and used in applications that do not have 3D accelerated graphic hardware, such as cell phones. Some renderings are shown in the picture below.






The methods for evaluation were systemic because they measure all of the system, from the 2D key views to the 2.5D interpolation model to the 3D rendering. The difficulty of 2D key views were evaluated and were found to be rather simple to make. Only 3 key view are needed and drawing the object is not that hard for an artist. The 2.5D interpolation was found to be comparatively easy to actual 3D interpolation and the final rendered 3D models were found to be plausible.

Discussion: 

This paper is interesting because it uses a not-so-familiar 2.5D interpolation model to render 3D objects. I took a graphics class last semester and it was fairly difficult dealing with 3D objects, so I can relate to how these 2.5D interpolation models can make things much simpler and more efficient (3D rendering requires lots of calculations). Plus, the fact that this method can be done in real-time makes and on non-accelerated 3D devices like cell phones makes it a lot better/more interesting. From what I have seen, this work is fairly new. From what I look up, only one other study has been performed using 2.5D models of buildings to render the final 3D construction. I can see this being used in the near future for things like cell phones and in flash player so that access to a 3D simulation can be viewed on devices with slower graphics hardware. I can also see it being used in cartoons and other various forms of art. I thought the idea was cool because I know how difficult it can be dealing with 3D objects, thanks to my graphics class.