Challenges

2D Symmetry Competition

Each competing team can enter any or all of the following FIVE types of symmetry detection challenges from 2D data sets :

1. Reflection; 2. Rotation

3. Translation 1 (frieze); 4. Translation 2 (wallpaper)

5. Skeleton Symmetry / Medial axis

1. and 2. Reflection and Rotation Symmetry Competition

Regular or near-regular forms are commonplace in our surroundings as well as in micro and macro-scaled worlds from molecules to galaxies. Discovery of such forms directly impacts our recognition, perception and (re)action in the real world. A recent survey indicates that in this critical aspect of cross-domain, cross-scale perception, computer vision has fallen largely behind human and animal vision, especially in robust real world near-regular pattern recognition.

a. Single ref/rot

b. Multiple ref/rot

c. Sym-COCO ref/rot

3. and 4. Translation Symmetry Competition

For Translation 1 (frieze) and Translation 2 (wallpaper), we use a dataset consisting of mixed, see-through, and building facade patterns.

Mixed

See-Through

Building Facades

5. Skeleton Symmetry (medial-axis) Competition

The following skeleton datasets will be used as the basis for the local symmetry (skeleton) detection challenge. Training and test sets from these benchmarks will be defined, and used as a basis for evaluation.

SK506 (subset of images from MS-COCO dataset).
WH-SYMMAX (symmetry axes on the Weizmann horse dataset)

3D Symmetry Competition

Each competing team can enter any or all of the following the following FOUR types of challenges from 3D data sets :

1. Global reflection symmetry (synthetic data)

2. Local reflection symmetry (synthetic data)

3. Global reflection symmetry (real data)

4. Local reflection symmetry (real data)

Google Sites

Report abuse