Unsupervised part representation by Flow Capsules

Abstract

Capsule networks aim to parse images into a hierarchy of objects, parts and relations. While promising, they remain limited by an inability to learn effective low level part descriptions. To address this issue we propose a way to learn primary capsule encoders that detect atomic parts from a single image. During training we exploit motion as a powerful perceptual cue for part definition, with an expressive decoder for part generation within a layered image model with occlusion. Experiments demonstrate robust part discovery in the presence of multiple objects, cluttered backgrounds, and occlusion. The part decoder infers the underlying shape masks, effectively filling in occluded regions of the detected shapes. We evaluate FlowCapsules on unsupervised part segmentation and unsupervised image classification.

References in corpus (12)

• Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
• SfM-Net: Learning of Structure and Motion from Video
• Multi-Object Representation Learning with Iterative Variational Inference
• Object-Centric Learning with Slot Attention
• Capsules for Object Segmentation
• MONet: Unsupervised Scene Decomposition and Representation
• On the Spectral Bias of Neural Networks
• Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions
• Sparse Unsupervised Capsules Generalize Better
• Efficient inference in occlusion-aware generative models of images
• Unsupervised Discovery of Parts, Structure, and Dynamics
• Geometric Capsule Autoencoders for 3D Point Clouds

Cited by in corpus (3)

• Contextual Guided Segmentation Framework for Semi-supervised Video Instance Segmentation
• Back to Square One: Superhuman Performance in Chutes and Ladders Through Deep Neural Networks and Tree Search
• Conditional Object-Centric Learning from Video