English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Embeddings are an important tool for the representation of word meaning.Their effectiveness rests on the distributional hypothesis: words that occur inthe same context carry similar semantic information. Here, we adapt thisapproach to index visual semantics in images of scenes. To this end, weformulate a distributional hypothesis for objects and scenes: Scenes thatcontain the same objects (object context) are semantically related. Similarly,objects that appear in the same spatial context (within a scene or subregionsof a scene) are semantically related. We develop two approaches for learningobject and scene embeddings from annotated images. In the first approach, weadapt LSA and Word2vec's Skipgram and CBOW models to generate two sets ofembeddings from object co-occurrences in whole images, one for objects and onefor scenes. The representational space spanned by these embeddings suggeststhat the distributional hypothesis holds for images. In an initial applicationof this approach, we show that our image-based embeddings improve sceneclassification models such as ResNet18 and VGG-11 (3.72\% improvement on Top5accuracy, 4.56\% improvement on Top1 accuracy). In the second approach, ratherthan analyzing whole images of scenes, we focus on co-occurrences of objectswithin subregions of an image. We illustrate that this method yields a sensiblehierarchical decomposition of a scene into collections of semantically relatedobjects. Overall, these results suggest that object and scene embeddings fromobject co-occurrences and spatial context yield semantically meaningfulrepresentations as well as computational improvements for downstreamapplications such as scene classification.

Deriving Visual Semantics from Spatial Context: An Adaptation of LSA and Word2Vec to generate Object and Scene Embeddings from Images