Contact
tyler.nowicki@gmail.com

CV
Summer 2017
PhD
Research in Traditional Animation

Modified Photon Mapping

Tyler Nowicki, William Cowan, and Steve Mann
More Info
University of Waterloo
Master's Thesis
Global Illumination and Approximating Reflectance in Real-Time

Modified Photon Mapping

Tyler Nowicki, D. Walton and D. Meek
More Info
University of Manitoba
Bachelor's Honours Thesis
Partitioning Trees for Fast 3D Rendering

Binary Space Partition - Image 3

Tyler Nowicki, D. Walton
More Info
University of Manitoba
Analysis of Speed in Traditional Animation
Normalized speed. Degree 2 polynomial fit of normalized speed. General Speed Model.
Value of the constant term of the degree-2 polynomial. Value of the linear term of the degree-2 polynomial. Value of the quadratic term of the degree-2 polynomial.
Data size analysis. Polynomial parameter degree analysis. Polynomial model degree analysis.
Example trajectory speed. Example trajectory speed. Example trajectory speed.
Tyler Nowicki
More Info
University of Waterloo
Chaos and Fractals
Generated by a choas game fractal. Generated by a choas game fractal. Generated by a choas game fractal.
Generated by iterated function system. Generated by iterated function system. Generated by iterated function system.
Generated by L-System with randomized rule selection. Generated by L-System with randomized rule selection. Generated by L-System with randomized rule selection.
Tyler Nowicki
More Info
University of Waterloo
Animation Stylization

Presentation on the papers "Computer-Assisted Animation of Line and Paint in Display's Paperman" and "Stylizing Animation by Example". These papers describe a technique for generating a motion field and using it to advect lines or paint that is applied by the animator. The animator supplies the key-frames and manually matches corresponding strokes. The strokes move with the vector field and change shape as needed between key-frames. Occlusions are handled by tracking the occluded points and advecting the hidden parts of the stroke accordingly. This approach allows a 3D animation to take on a wide range of styles, although it does regular key-frames to be manually painted. The number of key-frames that is required will depend on the complexity of the character and its changes in appearance. This technique was used to create the Disney short film Paperman.

I will not be posting the slides because they make use of images which I do not have permission to republish.

Tyler Nowicki
University of Waterloo
Change-Points in Traditionally Animated Movement

Classification of subsequences into accelerating (ease-out), decelerating (ease-in), and constant velocity (even).

Average speed of decelerating subsequences. Average speed of accelerating subsequences.
Decelerating to accelerating change-point. Accelerating to decelerating change-point.
Tyler Nowicki
More Info
University of Waterloo
Character Movement Descriptions and Notations

Presentation on movement description languages and notations including a review of methods used to specify and store movement on computer. Consider that Fred Astaire can effortlessly perform a complex sequence of whole body movements with many difficult poses. We often want to record the sequence of moves in the dance or capture the essence of a great performance? Animators try to describe the essence of a great performance using 10 animation principles (plus 2 for the drawing process). Pixar's John Lasseter argued that their animation tools allow animators to apply these principles while creating animation. Numerous methods in the research claim to reproduce the effect described by each principle however these claims are not backed up by research. Choreographers have developed many different methods for representing or recording movement including various vocabularies, notations, and precise descriptions.

Ballet is a movement vocabulary. That means that specific works refer to complex sets of actions, each of which has a specific movement associated with it. Similar approaches are used in the research including state machines, special purpose movement generators, and motion-motifs. Their results are enhanced using motion blending and motion texturing to make them seamless and give them variation. Unfortunately these techniques are limited to a finite set of precisely executed movements, making them difficult to apply to character animation in general.

Pose-to-pose notations such as Dancewriting depict the movement with a sequence of stick figures. The figures are very simple in appearance and are augmented with symbols to describe the 3D layout of the limbs and how the dancer should move through and between each pose such as speed and acceleration. Movement duration is specified in the unit of measures (music). Some interpretation is required by the dancer to complete the movement. Similar approaches are used in the research including method for interpolating boundaries, shapes, and skeletons between key-frames. There are many commercial uses of these methods, although the animator must tediously specify 3D coordinates because it is difficult to correctly infer depth from a 2D image.

Ideographic notations such as Labanotation, Laban Movement Analysis, Action Stroke, Motif Notation and others use sequences of symbols that combine to describe a complete movement over time. Several tracks of symbols, one for each part of the body, indicate the direction of motion, the space near the body that it travels through, its speed, acceleration, and other movement properties. The symbols are goal-oriented and thus must be interpreted by a dancer; unlike the position tracking used in computer animation that captures a precise description of movement. The symbols by themselves do not specify a specific movement. Laban Movement Analysis adds properties such as weight, flow, time and space. In the research methods for goal-oriented optimization interpret the abstract goals of a given movement to create the actual performance. However, such methods are limited to specific domains or actions and we have yet to see a *complete* notation for describing any movement.

The kind of precise recordings move movement used in computer animation were first developed to by choreographers, referred to as the Eschkol-Wachman Notation. This method records X,Y,Z positions of the salient parts of the body over time. Although this approach is not favored by choreographers because it is so tedious and difficult to manually record a movement this way it is used exclusively by computer animation. It suffers from the same problems in computer animation thus techniques such as inverse-kinematics and performance capture help ease the burden associated with this approach for recording or specifying movement.

We come back to the question of to capture the steps and essence of a performance. Precise techniques capture everything, but they probably capture too much, unfortunately current techniques for specifying movement do little to ease the burden. Vocabulary and pose-to-pose notations have been widely explored in the research and their limitations tested. Methods which use this approach can capture the steps of a performance but they seem to have difficulty describing the essence of a great performance. Little work has been done on ideographic notations or goal-oriented optimization that would describe how each part of the body moves and its qualities. A solution to the burden faced by animators today may be found in the ideographic notations used to describe complex movement in dance.

I will not be posting the slides because they make use of images which I do not have permission to republish.

Tyler Nowicki
University of Waterloo
Speed in Animation
Example trajectory, spacing-chart style interface and speed-model.
Example spacing-chart style interface and speed-model.
Tyler Nowicki
More Info
University of Waterloo
Animation and Vision
Example of trajectory in traditional aniation.
Example of trajectory in traditional aniation.
Example of trajectory in traditional aniation.

Presentation on how movement in animation is presented to and interpreted by the human visual system. Above are examples of several movement paths. Each sequence of dots, referred to as a sample, indicates the position of a character's head, foot or hand in a sequence of frames of traditional animation. The line connecting the dots depicts the sequence, it is not clear if we perceive linear movement something more complex between the samples. Apparent movement sometimes referred to as the phi phenomenon describes the impression of movement we perceive between two samples of animation. This is unlike the first-order movement detectors we have in our retinas. Although they detect movement due to the aperture problem they cannot determine the direction of the movement. Our visual system must use feature matching to determine direction of movement. This is referred to as second-order movement detection. Simple methods have demonstrated some success at following objects by matching visual similarity. However, these systems cannot reproduce the human visual system's ability to recognize a rotating object when it is depicted with a point-cloud. We can conclude that the visual system must perform vector decomposition over time and recognize objects by their 2D movement pattern. This is interesting from an animation perspective because it suggests that it might be possible to construct a kinematic system for specifying movement that satisfies the needs of our structure from motion visual system without requiring all of the detail required for a 3D animation. However, little research has been done on measuring the utility of the models used in animation packages or presented in the research.

I will not be posting the slides because they make use of images which I do not have permission to republish.

Tyler Nowicki
University of Waterloo
Interpreting Line Drawings
Identify surfaces by finding loops. Identify cuboids by finding connected surfaces.
Tyler Nowicki
More Info
University of Waterloo
Construction of 3D Cuboids from Line Drawings
2D Line Drawing of a Cube. Skewed Symmetry Hyperbola of Gradient Equation Rotated 3D Cube Constructed from Line Drawing
Tyler Nowicki
More Info
University of Waterloo
Video Decoding with FFmpeg
The decoder must be primed with packets of data that preceed the desired frame.
Tyler Nowicki
More Info
University of Waterloo
CGL Presentation
Automatic Inbetweening
Superposed pencil-test key-frames. 1 - Key-frame. 2 - Intermediate frame. 3 - Intermediate frame. 4 - Intermediate frame. 5 - Key-frame.
Tyler Nowicki
More Info
University of Waterloo
CGL Presentation
Movement in Animation
Movement of circular components between key-frames.
Tyler Nowicki
More Info
University of Waterloo
CGL Presentation
Animation and Pencil-Tests
Line of action. Gesture-pose. Cylindrical components. Rendered Details.
Tyler Nowicki
More Info
University of Waterloo
Hand-Drawn Presentation
Animated Construction of Line Drawings
Lines do not reveal the subject. The overall shape doesn't give it away either. We think we know how it will turn out. Then it surprises us with something funny.
Tyler Nowicki
More Info
University of Waterloo
Content-Based Image Retrieval
Original Image First Pass of Wavelet Encoding Second Pass of Wavelet Encoding Last Pass of Wavelet Encoding
Tyler Nowicki
More Info
University of Waterloo
Animated Mosaics
Frame 0. Frame 10. Frame 20. Frame 30.
Tyler Nowicki
More Info
University of Waterloo
Flower Patterns
Star patterns with a flower texture. Star patterns with a flower texture.

Star construction uses the polygons-in-contact approach to turn the transformed polygons into star patterns. As an artistic extension, the solid colour interiors of each star were replaced with flower images. Unfortunately images of flowers with petals that matched the points of each star were not available.

In gardens and in nature grasses, sticks, creeps and vines often emphasis boundaries and contours. To emphasis the boundaries between stars, the borders were similarly replaced with images of natural boundaries, in this case grasses. These images also were selected because their density was sufficient to obscure the boundaries of each star.

Tyler Nowicki
University of Waterloo
Celtic Knots for Valentines Day
Graph paper plan, points indicate open spaces. Celtic heart.

I extended the uniform lattice of the pleatwork described in Peter Cromwell's article "Celtic knotwork: mathematical art" to an arbitrary mesh. The mesh polygons can have any number of edges, and any convex shape. The mesh drawn on graph paper is used to plan the heart shape in the first image. Points indicate an open space and lines indicate an intersection of the pleatwork. Unfortunately, it wasn't quite ready for Valentines Day...

Tyler Nowicki
University of Waterloo
My Stippled Family
The original picture was taken in the mountains. The original picture was taken in the mountains.

Lloyd's method was iteratively applied to position stiples. A single iteration of Lloyd's method involves calculating the Voronoi diagram for a set of stipples, and determining the centroids of their corresponding Voronoi regions. The centroids are then used as the new stipple locations. This produces images that resemble an artistic style of painting as presented by Adrian Secord's paper "Weighted Voronoi Stippling".

Tyler Nowicki
University of Waterloo
Three Approaches for Collaborative Filtering Recommendation Systems
Analysis of Tagging on Flickr
Tyler Nowicki
More Info
University of Waterloo
Improved Light Model

Rendered with a new bump mapping method with nearest texture filter and an improved light model. A topographical map of the earth is the bump map.

Tyler Nowicki
University of Manitoba
Indirect Shadows

Indirect shadows are not usually so clearly delineated in real-life. The red light is sun light reflecting off the room's red carpet.

Tyler Nowicki
University of Manitoba
Modified Photon Mapping vs Photon Mapping

Modified Photon Mapping Photon Mapping
Modified Photon Mapping Photon Mapping

Modified photon mapping uses a different method for constructing radiance estimates than photon mapping. This reduces variance (patchyness) and also makes illumination in the scene easier to control.

Tyler Nowicki
University of Manitoba
Biquad vs Diffuse Approximation

Reflectivity Biquad Approx. Diffuse Approx. Negative Difference
0.8 - Low Biquad Diffuse Negative Difference
0.6 - MidLow Biquad Diffuse Negative Difference
0.4 - MidHigh Biquad Diffuse Negative Difference
0.2 - High Biquad Diffuse Negative Difference

These screenshots compare biquadratic and diffuse approximations for lightmaps. This scene, using the standard camera position and direction, isn't the best for viewing the difference between the techniques. However, a root-mean-squared (RMS) error analysis indicates that the biquadratic reduces the approximation error by about 50%.

The material is defined by the following formula. The reflectivity, 'a', corresponds to the first column of the table. The mirrior reflection vector is given by 'b' and the actual reflection vector is given by 'r'.

Reflectivity Equation

The result is modulated by the surface colour.

Tyler Nowicki
University of Manitoba
Modified Photon Mapping /wo Final Gather

Standard Photon mapping.
Standard Photon mapping.

First modification, improved sample collection.
First modification, new sample collection algorithm.

High quality rendering of first modification.
High quality rendering of first modification.

Second modification, fixed light values.
Second modification, fixed cosine (dot product) weighting.

High quality rendering of second modification.
High quality rendering of second modification.

Third modification, indirect-shadowing.
Third modification, fixed coordinate orders and added indirect-shadowing.

High quality rendering of third modification.
High quality rendering of third modification.

Fourth modification, fixed illumination values.
Fourth modification, fixed illumination values.
High quality rendering of fourth modification.
High quality rendering of fourth modification.

None of the images use final gather or importance sampling. OpenGL texture filtering is enabled on some of the images causing dark lines around the edges of the blocks.

The first image was rendered with standard photon mapping. The other seven images were rendered with a modified photon mapping algorithm. Improvements, corrections and fixes are made as well as low and high quality renderings of most changes.

Tyler Nowicki
University of Manitoba
Three Results

Direct illumination hits and pink reflection rays.

Rays traced from yellow light source toward blue plane. An intersection is marked with a green dot. A random reflection ray, at the point of intersection, is generated and drawn with a short pink line.

Global illumination of a simple scene.

Photon mapping without final gathering. Photons are traced from the yellow light source into the scene. The light source emission is attenuated to account for the distance each ray travels before an intersection. Each ray with sufficient intensity is reflected in a random direction, accounting for the noisey appearance. The colour of the photon at each intersection is recorded in a k-d tree. A light-map texture is generated for each surface using the radiating colour of the reflecting photons.

Texture map of the earth at night.

Rendered with direct illumination and intersection detection. The effect of light on the surface at sharp angles has been corrected.

Tyler Nowicki
University of Manitoba
Ray Trace Earth

Texture map of the earth at night.

The image appears rough because texture filtering was not used.

Tyler Nowicki
University of Manitoba
Simple Low-Distortion Texture Coordinate Generation

When generating texture coordinates on a height field the naive method will cause stretching and sheering. This research produced a technique for reducing shearing and stretching by taking into consideration the horizontal and vertical arc-lengths.

Tyler Nowicki
University of Manitoba
Feasibility of Watermarking Streaming Media

Watermarked Image

Tyler Nowicki
More Info
University of Manitoba
Enhancing BitTorrent's Peer Selection Algorithm
Torrent Network - Image 2
Sara McGrath, Tyler Nowicki, Lily Woo
More Info
University of Manitoba
Sensitivity Analysis of Discrete Quality Settings on the Performance of Doom 3
Tyler Nowicki
More Info
University of Manitoba