Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
H-SPAM generates hierarchical superpixels by iteratively merging regions from a fine initial partition, guided by deep features and object priors. The merging process is divided into two phases: preserving object consistency followed by controlled inter-object grouping, allowing for modulation via visual attention or user input. H-SPAM achieves state-of-the-art performance on standard benchmarks for hierarchical superpixel generation, outperforming existing methods in accuracy and regularity.
Hierarchical superpixels can now be generated with state-of-the-art accuracy and regularity, opening new avenues for multi-scale image analysis.
Superpixels offer a compact image representation by grouping pixels into coherent regions. Recent methods have reached a plateau in terms of segmentation accuracy by generating noisy superpixel shapes. Moreover, most existing approaches produce a single fixed-scale partition that limits their use in vision pipelines that would benefit multi-scale representations. In this work, we introduce H-SPAM (Hierarchical Superpixel Anything Model), a unified framework for generating accurate, regular, and perfectly nested hierarchical superpixels. Starting from a fine partition, guided by deep features and external object priors, H-SPAM constructs the hierarchy through a two-phase region merging process that first preserves object consistency and then allows controlled inter-object grouping. The hierarchy can also be modulated using visual attention maps or user input to preserve important regions longer in the hierarchy. Experiments on standard benchmarks show that H-SPAM strongly outperforms existing hierarchical methods in both accuracy and regularity, while performing on par with most recent state-of-the-art non-hierarchical methods. Code and pretrained models are available: https://github.com/waldo-j/hspam.