Skip to main content

Artistic Animation Rendering with Stylization

Periodic Reporting for period 1 - AniStyle (Artistic Animation Rendering with Stylization)

Reporting period: 2018-02-02 to 2020-02-01

"Traditional computer graphics renderings aim to make artificial images of simulated 3D environments that look 'just like the real world.' However, the objective of NPR is to make images in artistic visual styles for digital art. NPR tries to address the questions like ""What methods can we offer artists, scientists, engineers, and non-experts for creating artistic imagery?,"" ""How can we create artistic animations without unwanted perceptual artefacts?,"" or ""How can we evaluate the effectiveness of these methods, in terms of temporal coherence?""

Animation Stylization Techniques (ASTs) are widely used for special visual effect of movie, game, etc. Unfortunately, most animations stylized using current ASTs suffer from various artefacts such as flickering which can only be removed via laborious manual correction.
Three fundamental of temporal coherence:
1. Flatness - the perception that an image appears painted on planar canvas (flat medium) rather than on the surface of a 3D object, in order to convey the 2D aspect of a given stylization.
2. Motion Coherence - the high correlation between the motion of texture marks and the apparent motion of 3D scene. When the marks remain in same position within the scene instead of following the scene's motion, the shower door effect occurs - the marks seem sliding over the scene.
3. Temporal Continuity: this indicates that the texture marks are smoothly animated, without any sudden change which is seen as flickering. The goal is to keep the stylization as continuous as possible during the animation.

The aim of this project is to develop novel approaches for stylizing 3D models (AST-3D) and video sequence (AST-Video), addressing FMT problems simultaneously from two perspectives: (1) development of computer graphics and animation techniques; and (2) quantitative psychological evaluation of FMT, especially temporal coherence. To date, few researchers have assessed stylization effectiveness, and their techniques of aesthetic evaluation are crude and subjective. We propose instead to investigate visual attributes that underlie perceived temporal coherence and develop a mathematical model to evaluate the stylization effectiveness. The resulting improved understanding of the temporal coherence will support the further development of novel ASTs. Therefore the research objectives of this project are:

Evaluation of temporal coherence: We will use psychophysical methods to investigate key visual attributes affecting FMT and develop a mathematical model to evaluate the coherence. This will provide a precise quantitative measurement of the effectiveness of stylization techniques developed in next objectives.
Development of AST-3D: We will develop a texture-based method that not only satisfies FMT but also enables creation of new styles such as hedcut13 and mosaic for 3D animation.
Development of AST-Video: We will develop an example-based method to create various artistic styles (e.g. cartoon; painterly) for video. In fact, the method will have no limitation on style type and, like AST-3D, will satisfy FMT without need for labour-intensive manual corrections."
The project kick started in Feb 2018, and is supposed to finish by Feb 2020. Unfortunately, the project has to terminate earlier by the end of August 2018. Overall, the project has run smoothly and achieved the objectives specified in the Annex 1 of the Grant Agreement.

The project portal (website) has been setup for publicity and dissemination purpose, providing the public general knowledge about the project. The technical tasks such as mesh parameterization for AST-3D, view-dependent texture mapping, etc. have been developed, which are in line with the tasks specified in the work packages. A number of documents have been drafted and sent to the EU project portal, including the data management plan, the procedures and criteria for recruiting research participants, the information on the procedures for data collection, etc. We have also submitted a technical paper to the leading conference in the field to publicize the working progress and results, as a part of deliverables.
The project kick started in Feb 2018, and is supposed to finish by Feb 2020. Unfortunately, the project has to terminate earlier by the end of August 2018.