Unsupervised part representation by Flow Capsules
This work is significant for researchers working on unsupervised object and part representation learning, as it offers a novel approach to improve the part discovery capabilities of capsule networks.
This paper addresses the limitation of capsule networks in learning effective low-level part descriptions by proposing primary capsule encoders that detect atomic parts from a single image. By exploiting motion as a perceptual cue during training, the method achieves robust part discovery in challenging conditions and infers underlying shape masks, even for occluded regions.
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.