links

CNS/EE 148 - Spring 2000

Projects

Project presentations will take place on the 1st and 6th of June during nomral class hours. Graduating seniors and Graduate students should present on the 1st so we can submit grades in time. Please indicate to me if you have a scheduling preference. Presentations should be about 10 minutes with 5 minutes for questions.

Standard Project.

Extend detector so that it is invariant w.r.t translation, scale, and orientation.
- You will find this document quite usefull.
- Code for evaluating shape densities : Laguerre.m hist2D shape shapeDensityDM
- Remember to verify that you are indeed scale and rotation invariant by generating synthetic images with different orientations and scales (similar to hwk#5).
Implement speedup heuristics for hypothesis testing. Here are some hints. (fastsearch.ps)
- Code for doing breadth-first search breadth_first.m
Build interface for easy manual training
Evaluation of detector in searching for faces and cars, add one more database
cars.tar.gz noncars.tar.gz faces.tar.gz nonfaces.tar.gz
Class competition...make additional changes in order to be the one who can most quickly build a model of an object, most quickly search a database and retrieve images with the highest accuracy.

Systems for Object Model Learning

Ignore rotation and scale invariance
Automatic design of feature detectors from unlabeled images. Alternatively use generic detectors.
Automatic training of models from unlabeled images with clutter. see EM algorthm also cs156b
matlab code:

Time signals

Recognize writing: signatures or words using x(t),y(t) data gathered from pen tablet.
Recognize neuronal spikes: look at biologic data and identify particular patterns of neuronal firing. Wire a palm pilot directly into your cortex and demonstrate the ability to turn it on and off using only your thoughts.
Speech or other Audio: apply arrangement of parts model to features extracted from a stream of audio data. Look for characteristic blobs in the spectrogram so you can tell everytime someone says the world "Elvis".

Implement other recognzition approaches

Rowley: neural network face detector.
Schneiderman: statistical object detector.
Amit: photometery and location.

Take a gamble with the potential to win big.....

Build a system that can tell you whether an image contains an animal. Bonus Points: do it as fast as the human brain. (~150msec see Simon Thorpe, Nature 1996).
Build a system that can tell you what sort of room you are in (kitchen, bedroom, classroom, ...)

last modified 4/25/00