top of page

Research

Master thesis: Automatic Recovery of 3D Structure from an                              Architectural Scene Image

Anchor 1

Left: Original image.                        Middle: refinement orientation map.        Right: Inferred 3D structure with texture.

        Our thesis presents an automatic method to label different orientations of building façades in a single image and infer its 3D structure from these geometric labels. While most of the existing works use Manhattan World assumption, our method takes a more general assumption which agrees more than three vanishing points in an image. We firstly estimate the coarse orentation map from vanishing lines and propose a multi-cue optimization to find optimal orientation labels. Secondly, we segment the orientation map and recover the 3D scene according to physics-based criteria. We propose a physics-inspired objective function to evaluate the results of 3D structures. The preliminary experiments demonstrate that the proposed method can deal with complicated façades of architectures.

Term Projects

Anchor 2

3D Game Programming - Viking Uncle's fantasy drifting

This is a role playing and adventure game which made by Unity3D with Javascript and C#. The game story is that viking uncle met a shipwreck accident, he should survive on the island and build a raft to escape from island in limited time. 

Group members: YunChih Yang, ChienYing Lee

Computer vision - Combined Selective Search and Tracking-Learning-Detection

Our method mainly improved Selective Search [IJCV 2013]  for object detection and combined with Tracking-Learning-Detection [PAMI 2012]. To achieve a semi-automatic system, we added object detection to select the candidate for user. And we developed this project in linux system.

​Group members: YunChih Yang, YuDi Tseng

Computer Animation and Special Effects - A minions 3D animation

We produced a humorous 3D animation in the project which is about two minions have an argument. The prologue of this animation is like Pixar Animation. And we used the following techniques in Maya: Dynamic rigid body, Fluid, Human IK and Blend shape.

​Group members: Perry Lai, YunChih Yang

Image Based Modeling & Rendering - Improved Image Completion using Planar Structure Guidance

Image completion is a useful image editing tool, allowing users to remove unwanted objects in an image. We improved Image completion paper [SIGGRAPH 2014] by strengthening plane and extracting regularity. 

​Group members: Perry Lai, YuDi Tseng, YunChih Yang

Smart Phone Application Development - Hit Mosquito

We created a Hit mosquito app since mosquitoes in summer are annoying. We used Java Development kit and Android SDK in Eclipse IDE. The techniques about Bitmap, Canvas and Paint were also used in this project.

Group members: WeiHong Lee, YunChih Yang

Rendering Techniques and Applications - Crush Hit

Inspired by an app "Smash Hit", we produced a shooting game by C++ with OpenGL. The aim of this game is  to hit and smash everything.

Group members: YunChih Yang, YuSheng Lin

Please reload

Anchor 3
Anchor 4
Anchor 5
Anchor 6
Anchor 7
Anchor 8
pygame1.PNG
Python project - Pygame application

Motivated by a small game "NS-SHAFT", I developed a similiar game by pygame with python. The game purpose is to climb down as most floors as you can.

Link: https://repl.it/join/gpmfyxto-a18336833

Member: YunChih Yang

Anchor 9

Bachelor thesis: A Distortion-Sensitive Seam Carving for                                       Content-Aware Image Resizing Using Anti-                                 aliasing Algorithm

Group members: YunChih Yang, SzuYu Lin, ShanHuei Wang

(a) Original image                                                 (b) Scaling                                 (c) Cropping                    (d) Content-Aware Image Resizing

        Standard image scaling is not good since it is oblivious to the image content such as Figure (b) and (c). Our bachelor thesis was based on Seam carving algorithm [ACM Trans. Graph 2007]. We used saliency-based image attention model to detect the important objects in an image. To reduce the discontinuous segments of the image,  we considered Sobel edge detector to increase the energy of the  segments. Furthermore, we proposed to use Canny edge detector before applying anti-aliasing algorithm. Experimental results demonstrate the effectiveness and the efficiency of the proposed method.

bottom of page