The paper presents a model of visual object representation by fragment views, rather than canonically-oriented whole-object views used in Chorus systems. Following recent results (Sheinberg and Logothetis, 2001) on object representation in inferotemporal cells during free viewing, we implemented a simplified attentional system which yields fragment views of objects, which are then used to train object-tuned modules. Each object is represented by a complete RBF module, instantiating a representation space. We show that such a system can produce distributed representations, like Chorus of views systems, and that dissociating objects from retinotopy enables a fuller model of scene geometry analysis to be advanced.
|Title of host publication||Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02|
|Place of Publication||Singapore|
|Publisher||The Institute of Electrical and Electronics Engineers (IEEE)|
|Number of pages||5|
|Publication status||Published - 2002|