Introduction

Our goal is to start to get a handle on the variability and consistency of children’s everyday scenes and activity contexts, combining both the existing SAYcam annotations and analysis of the 397 scenes identified in the SUN dataset (used subsequently in the Places Database (Zhou et al. 2017)).

How did the SUN dataset build a somewhat comprehensive list of the environments that people experience? The SUN team selected categories with unique identities in discourse, using WordNet to select 70,000 words and concrete terms describing scenes, places and environments that could be used to complete the phrase “I am in ____”, or “let’s go to ____.” Most of the words referred to basic and entry-level names, resulting in a corpus of 900 different scene categories after bundling together synonyms, and separating classes described by the same word but referring to different environments (e.g. inside and outside views of a building).

To start identifying which scenes children are exposed to more often, we began by simply asking 7 adults (some of them parents) to rate each of the 1110 scenes for “How often could you find children here?” on a scale of 1 (Never) to 5 (Very often). Shown below is the distribution of mean scene relevance to children, as well as the most and least child-relevant scenes in the SUN / Places Database.

Distribution of scene relevance to children

Of the 1110 scenes, only a relatively small number of scenes are deemed to be experienced by children “Often” or “Very often”: there were 87 scenes (7.8%) with an average rating of at least 4.0. Moreover, a large portion of the scenes are expected to rarely (2) or never (1) be seen by children: 531 scenes (47.8%) fall in these categories.

Child-relevant Scenes

The 87 child-relevant SUN / Places Database scenes that children see often or very often, with standard error bars.

Child-unfriendly Scenes

The 50 least child-relevant SUN / Places Database scenes again seem very reasonable.

Combo graph

Scene Composition

## [1] 30

References

Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., and Torralba, A. (2017). Places: A 10 million Image Database for Scene Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence.