Semantic-guided modeling of spatial relation and object co-occurrence for indoor scene recognition

arXiv:2305.12661v3 Announce Type: replace
Abstract: Exploring the semantic context in scene images is essential for indoor scene recognition. However, due to the diverse intra-class spatial layouts and the coexisting inter-class objects, modeling contextual relationships to adapt various image characteristics is a great challenge. Existing contextual modeling methods for scene recognition exhibit two limitations: 1) They typically model only one kind of spatial relationship among objects within scenes in an artificially predefined manner, with limited exploration of diverse spatial layouts. 2) …

abstract adapt arxiv challenge class context cs.cv diverse however image images modeling object objects recognition relationships semantic spatial type