UC San Diego

Visual attention and memory research focus heavily on controlled experiments with simple visual features and discrete objects, despite evidence for an important distinction between representations of objects and large extended surfaces in scenes. Other work suggests that what we know about attention and memory for objects may not simply extend to scene-surface information, motivating us to better characterize these differences. Chapter 1 investigates a paradigm widely cited as demonstrating the existence of scene-specific representations in working memory, finding that it does not convincingly differentiate scene-specific information in working memory from information in high-capacity iconic memory, which could be in a number of different formats. This led us to explore other approaches to more specifically isolate scene surface information. For example, recent fMRI work has revealed a potential marker for scene-specific representations, which showed promise for investigating influences of different task or attention conditions. In Chapter 2, I tested whether this scene-specific information persisted when participants viewed naturalistic scene photographs rather than 3D-rendered environments. I used deep neural networks to estimate ground-truth 3D information about stimuli in a publicly available fMRI data set. Using this information to predict fMRI responses, I found evidence of 3D scene-specific representations, although this information was less distinguishable from 2D information than in the previous work. In Chapter 3, I re-examined this finding using stimulus photographs with ground-truth 3D information that, as a set, had more potential to differentiate 3D-surface features from 2D features. We also tested the presence of more spatially precise scene-specific information that could be more useful in moving through the 3D world, finding a shocking dominance of 2D visual information over both types of 3D information, with no evidence for scene-specific representations. This echoes behavioral work suggesting that 2D textures may underlie 3D representations in natural scene images and highlights the importance of studying complex real-world information using complementary stimulus sets that preserve different aspects of the natural world. Together, these chapters lay critical groundwork for understanding how scene representations behave under different attention and memory conditions.