Embed Size (px)
Citation preview
Stanford CS223B Computer Vision, Winter 2005
Final Project Presentations + Papers
Sebastian Thrun, Stanford
Rick Szeliski, Microsoft
Hendrik Dahlkamp and Dan Morris, Stanford
© sebastian thrun, CMU, 2000 2
Final Project Presentations
Tue March 8 P16: Road surface type estimation for
DARPA Grand Challenge P09: Finding Objects in 3d Point Clouds
of Urban Environments IP5: Tracking of Multiple RC Cars P15: Shadow detection and removal for
DARPA Grand Challenge P18: Gesture recognition for HCI
(Human-Car-Interaction) P04: Smilifying Images IP4: Analysis of Parking Patterns P07: Change detection from multiple
camera images P06: 3D SFM with an IMU P12: Shape Through Smog
Thu Mar 10 P11: Learning Optical Flow from Control
Commands of a Mobile Robot P03: Identifying Cell Components IP7: Mineral identification from MER
Pancam images P13: Aligning Images of Flies IP8: Real-Time Feature-Based
Mosaicking P17: Displaying high dynamic range
video IP3: Age Classification P10: Terrain and Object Matching in
Off-road Scenes IP6: Depth Estimation from Single
Images IP1: Single View Corridor
Reconstruction P02: Improved Speech Recognition
through Vision Localization P01: Autonomous Helicopter Pose and
Landing
© sebastian thrun, CMU, 2000 3
Project Presentations
Use MS Powerpoint Mail to [email protected] by 12am at the day of
your presentation• PPT file• All animations/videos (links please)
You have 6 minutes (sorry, this is a big class). WE WILL STRICTLY ENFORCE THE TIME
LIMITS
© sebastian thrun, CMU, 2000 4
Final Project Slides
Sorry this has to fit into 6 minutes
1 Slide with title + team member names 1 Slide with problem statement and data samples 1 slide with your approach (keep it short!) 2-3 slides with results, animations?
(hidden slide: list percentages of who in your team did what, e.g.: Dave did 80% of the work, Mike and Ron each 10%)
Example Presentation
(Dan Gindikin, CS223b 2004)
© sebastian thrun, CMU, 2000 6
Problem: Matching Images to Aerial Maps
T
z
y
x
R
'
'
'
z
y
x
© sebastian thrun, CMU, 2000 7
Approach: SIFT
Level 2 Level 3 Level 4
© sebastian thrun, CMU, 2000 8
Results7690features
968features
?
© sebastian thrun, CMU, 2000 9
Your Final Project Paper On-A-Slide Abstract (short is sweet!)
• Problem, gap, approach, key results Introduction
• Broad problem and impact• “scientific gap” (what technical aspects have not yet been solved)• summary approach (should include reference to technical gap) • key results
Approach• Background tutorial (if necessary)• Your technical innovation (might be multiple pages/sections, with repeated
reference to scientific gap) Results
• Main questions that are being investigated in experiments, ref to gap possibly with main results highlighted
• Data sets, simulator, implementation details• Empirical results (might be multiple pages)
Related Work• Don’t just say what’s been done. Point out how prior work relates to yours
and to the scientific gap you set forth in the intro. Summary/Discussions/Conclusion
• Summary problem, approach, result, in past tense• Discuss open questions, promising research directions
References
© sebastian thrun, CMU, 2000 10
Lesson # 1
Put yourself into the position of the reader!
© sebastian thrun, CMU, 2000 11
Lesson # 2
Motivate your problem• Why does it matter?• Why is it not solved yet?• What impact would a solution have?• What contribution did you make?
© sebastian thrun, CMU, 2000 12
Lesson # 3
It doesn’t matter how you got there
• “We tried A, it didn’t work, therefore we tried B”• “B works. To see, let us consider an obvious alternative
A, and show A does not work”
Document your progress, not just achievement• “B works”• “B improves over A (current techniques) by X, which is
important because of …”
© sebastian thrun, CMU, 2000 13
Lesson #4
Resist the temptation to say everything you know.
• A good paper makes one point, not two• A good paragraph makes one point, not two• (most points are only made in one paragraph, not too)
© sebastian thrun, CMU, 2000 14
Completeness and Conciseness
Reviewers may not be familiar with your area:• Provide Problem motivation• Describe Significant application domains• Introduce the State of the art/background material• Use Consistent Notation• Make sure your experiments match your claims• Describe and motivate your measures for evaluation
© sebastian thrun, CMU, 2000 15
Conference Reviewers are Overworked
• Don't expect them to pay attention to details• Don't expect them to read small fonts• Motivate problem, explain why open, why interesting• Present one idea, not two, three, ...• Pick informative title• A picture is worth 1000 words• Be concise! Get to the point!• Run a spell and grammar checker• Use terminology consistently• Define abbreviations, avoid them if possible• Convince reader that experiments fit claims/problem• Make sure the paper “flows”
