Overview
I Project LogisticsI Class Coverage and Ideas
I GeometryI Recognition
I Example ProjectsI Helpful Resources
Project Logistics
I Teams of 1-4: Number of people is taken into account whengrading project
I Suggestions for project directionI Replicate an interesting paperI Compare different methods to a benchmarkI Use a new approach to an existing problemI Implement an interesting systemI Original research
Sharing a Project with Another Class
I Sharing projects is generally allowedI Must be approved by both our staff and the other course staffI Project must be big enough that you can clarify which parts
of the project were done for which classI Each part must be substantial enough to hold as a single
projectI Will need a separate writeup for each class
Project Grading
I Course project is 38% of your final gradeI Project Proposal: RequiredI Midterm Progress Report: 5%I Presentation: 8%I Final Report: 25%
Project Proposal
I Maximum of 4 pagesI Submit the report as a PDF document through GradescopeI Include the following:
I Title and authorsI Sec. 1. Introduction: Problem you want to solve and whyI Sec. 2. Technical Approach: How do you propose to solve it?I Sec. 3. Milestones (dates and sub-goals)I References
I You will be assigned a project mentor
Project Midterm Progress Report
I Maximum of 4 pagesI Submit the report as a PDF document through GradescopeI Include the following:
I Title and authorsI Sec. 1. Introduction: Problem you want to solve and whyI Sec. 2. Technical Approach: How do you propose to solve it?I Sec. 3. Milestones achieved so farI Sec. 4. Remaining Milestones (dates and sub-goals)I References
Project Presentations
I Short presentation with time for a brief Q&AI Include the following:
I Problem Motivation/DescriptionI Technical ApproachI Results
Project Final Report
I Maximum of 10 pagesI Submit the report as a PDF document through GradescopeI Email your code to [email protected] Include the following:
I Title and authorsI AbstractI Sec. 1: IntroductionI Sec. 2: Previous workI Sec. 3: Technical ApproachI Sec. 4: ExperimentsI Sec. 5: ConclusionsI References
Class Coverage: Geometry
I Camera models and calibrationI Single camera and how we model it
I Single view metrologyI Estimating geometry from a single view
I Epipolar Geometry (Stereo Vision)I Estimating geometry from two viewpoints
I Structure from MotionI Using motion/several viewpoints to estimate structure
I Volumetric StereoI Using multiple views to map 3D points
Automatic Photo Pop-Up
Hoiem, D., Efros, A. A., and Herbert, M, “Automatic PhotoPop-Up”, SIGGRAPH 2005.
Scene Augmentation
Requires identifying the ball, players, referees, etc. in theimage/video in order to mimic occlusion
Photo Tourism
Snavely, N., Seitz, S. M., Szeliski, R. “Photo Tourism: ExploringPhoto Collections in 3D”, SIGGRAPH 2006.
Mobile Devices
Can you take an existing vision algorithm and adapt it to a mobiledevice to make it more useful?
Course Coverage: Recognition
I Fitting and matchingI Detectors and descriptorsI Object classificationI 2D/3D object detectionI 2D/3D scene understanding
Image SegmentationPartition an image into multiple segments (sets of pixels) in orderto make it easier to analyze
Image Completion
Hays, J and Efros, A. A., “Scene Completion Using Millions ofPhotographs”, SIGGRAPH 2007.
Face Detection
Detect the faces in an image: Used by Facebook for taggingimages and digital cameras for autofocus.
Other Topics
I Pose Estimation: Estimate the skeleton angles for a personfrom an image/video
I Action and Gesture Recognition: Is a person standing,walking, or sitting in an image/video? Is he/she waving?
I Scene Understanding: Can you classify a scene? Can yourecognize and/or segment each component of the scene?
Where to get Project Ideas
I Course Staff: Posted on website and/or PiazzaI Computer vision papersI Computer vision research groups at Stanford
I Silvio SavareseI Fei-Fei Li
I Last year’s projects: See course websiteI Come up with your own!
Datasets
I Many are available on the webI See the following aggregators:
I CV Datasets on the WebI Yet Another Computer Vision Index To Datasets (YACVID)I Computer Vision Datasets
Project Advice
I Choose your team wellI Make sure the scope of your project fits a quarter
I Set a minimum goal, desired goal, and a moonshotI Constrain your problem smartlyI See what datasets are available if you are doing a recognition
projectI You may need to plan ahead/learn outside materialsI Use software when available: OpenCV, PCL, MATLABI Come ask questions – We’re happy to talk!