Projects proposed to the students

• New data sets:

Collect new data sets and train and test object models on those. Suggested data sets:
• Fingerprints: develop a minutia detector and use it for fingerprint verification (one-to-one matching). (*)
• Spike trains from neurons: given the train of spikes produced by a neuron identify the corresponding stimulus.
• Signature recognition: develop feature detectors for handwriting data and identify people from their signature.
• Cursive handwriting recognition: develop f.d. for handwriting and detect letters and words in handwriting.
• Side-views of cars. Develop the detectors and recognize cars in traffic. (*)
• Efficient search algorithms:

Given a probabilistic feature-geometry model containing many features ranking all hypotheses is time-consuming. Develop an algorithm which searches the set of hyposthesis efficiently, prioritizing the more likely hypotheses. (*)
• Model learning:

Implement a given model learning method which automatically selects discriminative features and estimates the statistics of the corresponding constellation model. See paper by M. Weber et al.
• New feature detectors:

Implement feature detectors based on neural networks. See papers by:
• Rowley et al., PAMI 1998.
• Schneiderman and Kanade, CVPR98.
• Multiple object models:

Train constellation models for several different object classes and come up with ideas to efficiently detect objects of all classes in a given image. (*)
 

(*) Challenging.  You need to schedule a session with the TA in order to develop a careful plan.