COS 429 - Computer Vision

Administrative information

       Final project prelim submission (12/12)

       Final project submissoin and write-up due (1/13/09)

Project decription

The project is based on a Nintendo Wii game "Finding Miis", one of the games from "Wii Play". You will be given video excerpts from the gamplay and will be required to implement algorithms which accomplish certain tasks, which are described below (and they match the tasks one would have to accomplish during the gameplay). Moreover, within each task there are multiple "levels", which are sorted by their complexity. You are not required to complete all the levels in all the tasks; for more details on this, see the evaluation section below.

Tasks:

• Find look-alikes: In this task you will be given a scene containing a lot of Miis. A pre-set number of them (2 to 5) will be identical, and your job is to recognize the identical ones.
• Find the odd one(s) out: In this task, one or more of the Miis will be "different" than the other ones. Your job is to recognize which one(s) are different than the rest (the feature that is different will be given to you in the task description)
• Find the sleepyhead: In this task, one of the Miis in the scene keeps falling asleep. Your job is to find that sleepy one.
• Find the given Mii: In this task, you are given an image of a certain Mii beforehand, and are asked to find that Mii in the scene.
• Find the fastest Mii: In this task, one of the Miis in the scene is faster than the other ones in either walking, running, or swimming. Your job is to locate that Mii.

Levels:

Within each task, there are three levels: easy, medium and hard. This sorting may be somewhat arbitrary, as it is based on our subjective assessment, performance of our algorithms, and the hardness of the level in the original Nintendo Wii game. Of course, none of these can be completely accurate, so do not be confused if you find a harder level is easier for you. The factors that contribute most to the hardness of a level are the number of Miis in the scene, the number of Miis needed to finish the task, the background of the scene they're set in, and their type of motion.

Tasks and levels

Evaluation

Evaluation of your performance will be based solely on the successfulness in implementing correct fast algorithms to complete the above tasks. The evaluation has three aspects to it:

• Scoring a level: Your score for a given level of a given task is calculated according to a pre-defined formula (accuracy*fastest_time/your_time) , where accuracy is a Boolean variable which takes value 0 if you did not successfully accomplish the task and 1 if you did, fastest_time is the CPU time that it took for the fastest algorithm in the class to come up with the solution to the task, and your_time is the CPU time that it took for your algorithm to come up with the solution to the task. The task is successfully accomplished if the area of overlap between the predicted bounding box of the face and ground truth bounding box of the face exceeds by the formula:

  

• Minimal requirement: You have to successfully complete at least 1 level from at least 3 different tasks. This will mean probably finishing the "Easy" level in three different tasks.
• Competition: For this part, each team will get one final score. This score is a sum of weighted scores on all tasks. The score for each particular task is the highest weighted score on all levels from that task. That means, if your weighted score on the "Medium" level is higher than your weighted score on the "Hard" level, we will count your "Medium" score even though you correctly solved "Hard". The original scores on "Easy", "Medium", and "Hard" levels are weighted by 1, 1.5, and 2 respectively to get the weighted score. As a result, the maximum possible score for the Hard level is 2 times higher than the score for the Easy level, and the the maximum possible score for the Medium level is 1.5 times higher than the score for the Easy level.

Useful links

Skeleton Code

OpenCV Library Wiki

OpenCV on Ubuntu Linux

Mac OS X OpenCV Port

OpenCV Yahoo group

OpenCV windows package