This page contains the shape bias stimuli sets used in a study accepted at the International Conference on Machine Learning. Here is the citation for the paper:
Ritter, S*., Barrett, D. G*., Santoro, A., & Botvinick, M. M. (2017). Cognitive psychology for deep neural networks: A shape bias case study. In International conference on machine learning (pp. 2940-2949). PMLR. (*equal contribution)
In this study, there are 10 exemplar sets. Each set contains a target object, a shape match and a color match. These are shown in folder called 10 SETS. THESE ARE NOT THE IMAGES USED BY RITTER, BARRETT, ET AL, but merely show the 3 objects in each set. The images, each containing 1 object, are in the folders labelled by background context.
One of the most fundamental questions in psychology concerns the role of experience. What are the essential components of human experience? What are the malleable points in which small differences in experience can lead to different developmental outcomes? What are the mechanisms that underlie developmental change?
Answering all these questions requires that we know much more than we do about everyday experience. With these larger questions in mind, the Homeview project is collecting a large corpus of infant perspective scenes (using head cameras) and audio in the home as infants 1 to 24 months of age go about their daily life.
The corpus, with over 500 hours of head camera video promises new insights into the natural statistics of visual experiences for visual development generally, for visual object recognition, for human face perception, and for object name learning. We extract images from the video at 1 per second creating an image corpus of nearly 2 million images.
Participants were 91 infants (46 female, 45 male) aged 1 to 24 months from middle class families in Monroe County, Indiana who were recruited through county birth records and community events.
We have an additional 40 participants (24 female, 16 male) aged 1 to 15 months from a fisherman community in Chennai, India.
Recording the availability of faces in infants' everyday environments requires a method that is not disruptive of those daily environments. Accordingly, we use a wearable camera that was lightweight, cable-free, attached to daily-wear hats, and easy for parents to use.
In a pre-visit, parents were informed about the goal of the study, consent was obtained, and they were instructed to use the camera. A hat was selected and fit to the child. Subsequently, the materials were delivered to the infant's home and the parents were reinstructed in the use of the camera.
Parents were not told that we were interested in faces or social events but were told that we were interested in visual development and the typical range of visual experiences of their infant. They were asked to record during the infants' waking hours and to try to capture four to six hours of video during daily activities when the infant was awake and alert.
Because of the complexity and demands of parenting young infants, they were given up to two weeks to complete their recording. The cameras were collected once parents had completed their recording; parents were debriefed and consistent with the consent procedure were asked if they wanted any segments deleted.
The videos collected from parents were screened for privacy and accidental recordings (1.5% of total recording) and those sections were subsequently removed from the dataset.
Trained coders used one of two techniques to answer specific research questions about infant experiences like, "What proportion of the recording contained footage of human faces?" or, "What do the visual scenes of mealtimes look like?"
Continuous video: Coders would watch the videos and annotate segments of interest. These segments would in turn be coded for more specific questions.
Down-sampled images: Videos were converted to images which were in turn selected at specific time intervals (typically one image selected from every 5 seconds of video). Coders would annotate each image to answer specific questions.
The corpus coding is an ongoing process and changes with each major research question that we ask. Research papers based on this project contain detailed descriptions of how we coded for each research questions.
In the Lab
Parent and Toddler Free Play
In the Home
These videos were obtained in the home setting. See how selective our momentary view of the world is and how it changes with development.
