Public articles linked to the same research event.
arXiv The work formulates slot attention as an interacting particle system and derives simplified slot attention (SSA), a parameter-free variant connected to soft K-means, soft spherical K-means, and a derived unnormalised surrogate USSA; on Pascal VOC 2012 with frozen DINO features, SSA achieves segmentation and reconstruction performance close to full slot attention, suggesting its competitive assignment dynamics already account for much of the object-centric clustering while learned components mainly refine representations.
The work formulates slot attention as an interacting particle system and derives simplified slot attention (SSA), a parameter-free variant connected to soft K-means, soft spherical K-means, and a derived unnormalised surrogate USSA; on Pascal VOC 2012 with frozen DINO features, SSA achieves segmentation and reconstruction performance close to full slot attention, suggesting its competitive assignment dynamics already account for much of the object-centric clustering while learned components mainly refine representations.
The work formulates slot attention as an interacting particle system and derives simplified slot attention (SSA), a parameter-free variant connected to soft K-means, soft spherical K-means, and a derived unnormalised surrogate USSA; on Pascal VOC 2012 with frozen DINO features, SSA achieves segmentation and reconstruction performance close to full slot attention, suggesting its competitive assignment dynamics already account for much of the object-centric clustering while learned components mainly refine representations.
The work formulates slot attention as an interacting particle system and derives simplified slot attention (SSA), a parameter-free variant connected to soft K-means, soft spherical K-means, and a derived unnormalised surrogate USSA; on Pascal VOC 2012 with frozen DINO features, SSA achieves segmentation and reconstruction performance close to full slot attention, suggesting its competitive assignment dynamics already account for much of the object-centric clustering while learned components mainly refine representations.