In this paper, we propose a novel hierarchical image completion approach using regularity statistics, considering structure features.

This work explores the problem of generating fantastic special-effects for the typography. our key idea is to exploit the analytics on the high regularity of the texture spatial distribution for text effects to guide the synthesis process.

Implemented SIFT-based and deep-based image detection algorithm.

---

System Functions

• View dataset by category
• View detection results and the groundtruth
• Single image detection and batch detection
• Caculate mean average precision
• Three different algorithms to select

Algorithms

• SIFT-based: SIFT features are matched between the test image and training images to estimate affine transformations. The masks of training images are warped onto the test image and then accumulated to form a confidence map. The logo is extracted using the region growing algorithm over the confidence map.
• Multi-Task Deep Networks: Images are divided into K*K grids. Deep networks based on Inception-BN is used to classify each grid.
• Fast R-CNN: Implemented the paper: Ross Girshick. Fast R-CNN. ICCV 2015.

Detection Results

We tested on the FlickrLogos-32 dataset. The MAPs of our three algorithms are as follows:

We have also analyzed their performances in six respects:

Implemented the paper: Nister, David, Stewenius, Henrik. Scalable Recognition with a Vocabulary Tree. 2006:2161-2168.

---

System Functions

• Extract SIFT Features
• Build Vocabulary Tree
• Quantify Images as Vectors
• Find the most Similar Images

Here are some query results of our system. The first column is the query image and the last three columns are the corresponding query results.

Implemented a video processing system that provides vide coding/decoding, stabilization and denoising functions.

---

Video Codec

• Use ffmpeg to decode every frame from the video
• Convert the AVFrame to IplImage, then stabilize or denoise the image
• Convert the handled IplImage to AVFrame then write to a raw video
• Encode the raw video

Video Stabilization

• A pyramidal version of Lucas-Kanade optical flow computation
• A six-parameters affine transformation model
• A Gaussian motion smoothing

Denoising

• Non-Local Means filter

e-Course is a mobile application helping Peking University students to select, comment and search courses. *I tried to make my PPT professional*

---

It is common for students to share course information with each other using online social networks (OSN). However, it is difficult to organize and categorize the information posted by individuals in an ad hoc way on OSN. Our application provides students a venue to publish their ratings to every course they have taken in an organized way such that others can find useful course information from experienced students in an easy and reliable way.

Project Presentation

Click the picture below to see the project presentation:

We learned MIPS and how CPU ran machine codes in pipeline this semester. We were asked to use C++ to simulate a CPU.

---

Our program reads in assembler codes(which will then translated into machine codes) or machine codes and simulates how a CPU runs them. We provide CPUs with and without pipeline. In pipeline mode, our program handles data hazard and control hazard. The MFC GUI makes our program easy for users to understand the computer operational principle.

CPU Simulator without Pipeline

CPU Simulator with Pipeline

Our team members like to play the music game OSU!. Therefore we developed our own Taiko.

---

Game User Interface

The drumbeats come out from both left side and right side, with characters (A, S, ;, ') on them. Players shall push corresponding buttons when drumbeats moving into the centre circle.

Evaluation System

We evaluate players' performances according to valid drum beats, the precision and combos. We have five ranks: S, A, B, C, D.

My first attempt at an MFC GUI. It's much more difficult than it looks :)

---

Our team endeavoured to give users good experiences. Thus we learned MFC and designed the interface by ourselves. Here is the stage introduction:

STAGE ONE

This stage contains EECS mascots designed by Shuai Yang and Yiqing Xue. They looks cute but the game may not be as easy as you think.

STAGE TWO

This stage contains similar characters, which needs more time and concentration to differentiate. Note that even the same characters may have different fonts :-P

STAGE THREE

This stage is the extension of Stage Two. You need click the picture to turn over it and get characters. After this stage, we will embrace the ultimate stage with extraordinary difficulty.

STAGE FOUR

……, ……, ……, well well ╮(￣▽￣")╭

It's really a challenge for me to come up with an AI smart enough to battle with human after only half a year's study of programming.

---

In this project, though knowing nothing about GUI, I tried to make the character-based interface user-friendly. Here is the start screen:

And this is the game user interface:

I have also added winning and losing pictures using ASCII art.