Minimal-Cut Model Composition
Abstract
Constructing new, complex models is often
done by re-using parts of existing models, typically by applying a
sequence of segmentation, alignment and composition
operations. Segmentation, either manual or automatic, is rarely
adequate for this task, since it is applied to each model
independently, leaving it to the user to trim the models and determine
where to connect them. In this paper we propose a new composition tool.
Our tool obtains as input two models, aligned either manually or
automatically, and a small set of
constraints indicating which portions of the two models should be
preserved in the final output. It then automatically negotiates the
best location to connect the models, trimming and stitching them as
required to produce a seamless result. We offer a method based on the
graph theoretic minimal cut as a means of implementing this new tool.
We describe a system intended for both expert and novice users,
allowing easy and flexible control over the composition result. In
addition, we show our
method to be well suited for a variety of model processing applications
such as model repair, hole filling, and piecewise rigid
deformations.
Full paper : SMI05_MinCutModelComposition.pdf
(2,503kb). BibTeX.
Some Results
More results can be found in the paper.
The Centaur

A typical composition session. The stages involved in creating the centaur model from those of a
man and horse. (I) Input. (II) Placement
(semi-automatic, using our novel part-in-whole alignment method, or
manual). (III) Constraint selection (manual).
(IV) transition volume selection (manual
or taking the bounding box of the union as a default selection). (V)
The recovered transition
surface
(automatic). (VI) Clipped models (automatic).
The man model, courtesy of Cyberware
and Headus.

The final, stitched, centaur
result.
Composing Heads
Reversing constraints.
From left to right: The two busts; The models overlayed; First example,
taking the blue face and red head; Second example, reversing
constraints, now choosing the red face and blue head. In both, top is
the cut result (clipped models) bottom is the final result.
Both bust models, courtesy of Cyberware
Cerberus

Multi-step composition.
Cerberus, the mythical guardian of the gates of hell, is created here
in four steps. The top row shows overlays of the models used to create
the result in each step (shown in the bottom row). In step 1 we deform
the dog's head, turning it to make room for the other heads. Additional
heads were created by cloning and rotating
the whole dog. In each step we constrained the new dog's head and the
existing dog's torso and nose(s). The tail was added by composing the
existing tail
with a serpent model (in blue).
Model Restoration
Model restoration. Fixing
the scars and broken nose on the Igea artifact model, in three steps.
(a) Input model and the user drawn boxes around the flaws. (b) Overlay
of the input model and the aligned database bust model chosen to fix
each flaw. (c) Clipped models. (d) Input (left) and result (right).
The Igea artifact model,
courtesy of Cyberware.
Model Hole Filling
Hole filling. Artificial
holes opened in a bust model by removing both the nose and the top of
the head. These were automatically repaired in two steps. In each row
on the left is the input model (with the user drawn box around
the hole) and on the right is the final result (the little lump on the
man's head is not an
artifact. It is the tip of the cap worn by the scanned subjects in the
database).
Bust models used for model restoration and holefilling
are free samples from the CAESAR
database.
Chair
Composing a chair. From
top to bottom, the three composition steps in creating a chair. On the
left are the models overlayed; on the right are the composition
results. Note that in each step, multiple models (e.g., the four legs,
the six back rests) were treated as one model and composed in a single
step with an additional model (e.g., the seat).
The final chair result.
To
the Weizmann Institute of Science Home Page
Faculty
of Mathematics and Computer
Science
Computer
Vision Group
Last update:
Dec.
08,
2005
For
questions and remarks please contact:
Tal Hassner.
Confidential - Proprietary
Information.