Learning interpretable object and attribute features from images
LinSlot: Exploiting Linear Representation hypothesis for unsupervised attribute discovery from slot based object representation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
This research focuses on how to teach computers to recognize objects and their different traits like color or shape from pictures without any guidance. The authors improve existing methods by assuming that object traits can be represented by combining simple building blocks in a straight line, making the computer's understanding clearer and easier to use. Their approach helps the computer break down images into meaningful parts and traits, which can also be edited or manipulated. Experiments show their method works better than others in capturing distinct object features and traits.
What this means in practice
- •For computer vision engineers: Improve image editing tools by enabling fine-grained control over object attributes using interpretable representations.
- •For graphics software developers: Develop features that automatically detect and manipulate object parts and attributes in images for creative content generation.
Authors
Sanket Gandhi, Utkarsh Giri, Varun Subramanium, Rohan Paul, Parag Singla
Abstract
This paper studies the problem of learning disentangled representations of objects and their attributes from raw, unstructured image data. Slot-based methods have shown considerable success in unsupervised learning of object representations from images. Block-slot attention-based methods extend this framework to attribute representations by assuming a uniform factorization of object representations into attributes, which may be suboptimal and consequently limit the quality of the learned representations. We therefore investigate a framework for jointly discovering object and attribute representations. Our key contribution is leveraging the Linear Representation Hypothesis (LRH), which postulates that composable concepts can be represented as linearly additive subspaces in slot representations. Based on this insight, we propose a probabilistic model connecting images, slots (objects), and blocks (attributes). We present an architecture that leverages block attention to connect attribute representations to slots and incorporates LRH in both object and attribute representation spaces. This architecture effectively optimizes the Evidence Lower Bound (ELBO) of the proposed graphical model. Our experiments demonstrate (i) effective discovery of disentangled object and attribute representations, (ii) empirical evidence for LRH in slot space, and (iii) the ability to perform image editing owing to the disentangled and interpretable nature of the learned representations. Our experiments on multiple datasets demonstrate improvements in DCI scores over state-of-the-art methods.