Bibliography:
Tracy Hammond and Randall Davis. Tahuti: A Geometrical Sketch Recognition System for UML Class Diagrams. SIGGRAPH '06. 2006.
Summary:
The paper discusses Tahuti, a system that recognizes UML diagrams and thereby helps user sketches to be leverages in actual coding. There were a few systems such as Rational Rose, Ideogramic UML for recognizing UML diagrams, but were not very flexible in combining user sketches and recognition techniques. Other systems such as onces based on Rubine features required training.
Tahuti is a multi-layer framework for sketch recognition that leverages geometric properties of the sketches to recognize them. The system has four main steps,names preprocessing, selection, recognition, identification. Preprocessing produces a bunch of possible interpretations of the stroke data along with its probability. Strokes are grouped into collections (and also pruned) by using temporal and spatial data. Experimentally, it was found that 9 strokes is a good threshold for a collection of strokes. The Recognizer interprets strokes as either an 'viewable object' or an editing command. The eight possible viewable objects are general class, interface class, inheritance association, aggregate association, dependency association, interface association text or collection of unrecognized strokes. The identification stage selects the final interpretation based on a set of rules, that accounts for object movement, number of strokes and correctness probability.
Experiments showed that users preferred Tahuti over Paint programs and UML tools such as Rational Rose as it combined the ease of drawing with ease of editing UML.
Discussion:
Though this paper was written in 2002, it still seems the state of the art in UML recognition tools! Having worked with UML diagrams and white board drawing, I would love to use a tool like Tahuti for design discussions in a team. I wonder how one can coming this with gesture recognition techniques and maybe create a software project scaffold just from UML? That would be interesting...
Tracy Hammond and Randall Davis. Tahuti: A Geometrical Sketch Recognition System for UML Class Diagrams. SIGGRAPH '06. 2006.
Summary:
The paper discusses Tahuti, a system that recognizes UML diagrams and thereby helps user sketches to be leverages in actual coding. There were a few systems such as Rational Rose, Ideogramic UML for recognizing UML diagrams, but were not very flexible in combining user sketches and recognition techniques. Other systems such as onces based on Rubine features required training.
Tahuti is a multi-layer framework for sketch recognition that leverages geometric properties of the sketches to recognize them. The system has four main steps,names preprocessing, selection, recognition, identification. Preprocessing produces a bunch of possible interpretations of the stroke data along with its probability. Strokes are grouped into collections (and also pruned) by using temporal and spatial data. Experimentally, it was found that 9 strokes is a good threshold for a collection of strokes. The Recognizer interprets strokes as either an 'viewable object' or an editing command. The eight possible viewable objects are general class, interface class, inheritance association, aggregate association, dependency association, interface association text or collection of unrecognized strokes. The identification stage selects the final interpretation based on a set of rules, that accounts for object movement, number of strokes and correctness probability.
Experiments showed that users preferred Tahuti over Paint programs and UML tools such as Rational Rose as it combined the ease of drawing with ease of editing UML.
Discussion:
Though this paper was written in 2002, it still seems the state of the art in UML recognition tools! Having worked with UML diagrams and white board drawing, I would love to use a tool like Tahuti for design discussions in a team. I wonder how one can coming this with gesture recognition techniques and maybe create a software project scaffold just from UML? That would be interesting...
No comments:
Post a Comment