Announcement
How vision transformers group image patches into objects
A researcher says new work with a colleague draws on Gestalt psychology to study how vision transformers segment objects.
TLDR
A researcher says vision transformers can learn to segment objects. In new work with a colleague, they draw on Gestalt psychology to investigate how the models bind image patches into objects, comparing it to a question psychologists asked about people a century ago.
Combined views
18.8K
3 Sources, first seen ago
