PixNet: A Localized Feature Representation for Classification and Visual Search

Weakly Supervised Graph Based Semantic Segmentation by Learning Communities of Image-Parts

“…Inspired by [37,48], we introduce an image-driven graph to capture both visual and spatial characteristics of image regions. This is followed by a graph partitioning algorithm to segregate highly related regions across all images in the database.…”

“…To determine the association between each of its segmented regions and the detected communities, we follow [37] and provide a brief summary. Let H i denote the set of all nodes in the spatial neighborhood of node i, c j be a community with j ∈ {1, .…”

Retrieval of Images with Objects of Specific Size, Location, and Spatial Configuration

“…In this section we follow the localized feature representation introduced by [20] and provide a brief summary. We define a network of segmented image regions to integrate the visual similarity between segmented regions across all training images with the localized spatial information.…”

Search and retrieval of multi-modal data associated with image-parts

“…Following the work of [21,22], we integrate the visual characteristics along with the spatial information of image-parts across all database images in a graph structure. To provide spatial information, we utilize a segmentation algorithm based on color and texture [23].…”

“…For example, in the graphical image structure, each community contains all the pieces/image-parts of an object and mapping these back to each segmented image would highlight/detect that particular object [21,22]. A detected community for the textual tags' graph corresponds to highly related/correlated tags.…”