15-869 B Administrative Information for Spring 2014
Grading for the class will be as follows:
- 10% Class participation
- 20% Blog / web page
- 20% Paper presentation
- 15% Project proposal
- 35% Final project
Notes for the Paper Presentation
This section has just a few notes on what I am expecting for your
You will have 30 to 40 minutes to present your paper. Aim to fall
within this timeframe including discussion and questions (which you
should encourage). Thirty to forty minutes is a lot of time. It is
significantly more time than the authors had to present their papers
at whatever conference they may have appeared. You have time to
develop some amount of technical detail. However, you probably cannot
cover all of the detail in the paper, so you have to make some careful
choices about what to talk about and what not to talk about. If you
are unsure about how much detail you can hope to cover, first devise a
tentative plan and then come talk to me.
I expect your talk to go generally like this. Depending on the topic
of the paper, the approach of the authors, and your own personal
style, you will probably have to adapt my guideline below. In
particular, I've put some possible timing information just as a rough
guideline, but this will vary a lot from paper to paper and
presentation to presentation.
- (2 minutes) Where did this paper come from? Where was it
presented, who are the authors, what inspired you to have a look at
- (2 minutes) What is the overall big picture motivation of the
authors? What seems to be driving them to pursue this research?
- (1 minute) What did they actually achieve? You can show part of
the video here to give people some idea of what to expect.
- (5 minutes) Now, you can give some background. This will come
from the paper's related work section. Probably a whole lot of other
people have tackled this problem. If you'd like, you can briefly show
some of previous work. Then, try to state the precise contribution of
this paper beyond previous work in your own words. If it is a really
top quality paper, the authors will have made this easy for you.
- (5 minutes) Now we are heading into the core part of the paper.
You can start by giving the problem definition very precisely. Think
inputs, outputs, system overview diagram, whatever will help us
understand the exact way in which the authors think about this
- (10 minutes) Now that we understand the problem statement, tell
us how they solved it. Try to do this in a top-down way, focusing on
the most interesting and important parts and only introducing detail
as necessary. Use lots of pictures and diagrams here.
- (5 minutes) Onto the results. How did the authors validate their
approach? Did they compare against alternatives? Did they simply
show videos? Did they test for statistical significance? Did they
compare against the real world? Show the key outcomes of the
algorithm or study. You may want to segment up their video and/or
separate out their results figures into subsections to help you better
present it, or you may find it works ok to show everything at once.
- (5 minutes) Now discuss. The authors will have their own
discussion points, but make this section your own. Did the authors
achieve what they originally set out to do? If you could summarize
this paper in just a few words, what would they be? What is good and
bad about the approach? What questions were raised in your mind?
What limitations did you see, and what opportunities for future work?
- Make sure you invite questions, not just now at the end, but
along the way as well.
The SMARTBODY system
The DANCE software system
To my knowledge, all three of these projects use the Open Dynamics Engine
Another simulator that we have been using recently is Bullet.
There are a number of other academic based simulation engines you can play with, including RTQL8 developed by Karen Liu's team at Georgia Tech, MuJoCo, developed by Emanuel Todorov's group at the University of Washington, and Klamp't, develped by Kris Hauser at Indiana University. More are showing up all the time, so you may find yet another system that looks promising!
The CMU motion capture database
is a good source for motion capture data
The software tools on
the CMU website may be useful for viewing motion capture data and
making kinematic edits, but do not have an integrated physics engine.
Doing so would be a substantial project.