Multi-Object Representation Learning with Iterative Variational - - PowerPoint PPT Presentation

multi object representation learning with iterative
SMART_READER_LITE
LIVE PREVIEW

Multi-Object Representation Learning with Iterative Variational - - PowerPoint PPT Presentation

Multi-Object Representation Learning with Iterative Variational Inference Klaus Greff, Raphal Lopez Kaufmann, Rishabh Kabra, Nick Watters, Chris Burgess, Daniel Zoran, Loic Matthey, Matthew Botvinick, Alexander Lerchner Multi-Object


slide-1
SLIDE 1

Multi-Object Representation Learning with Iterative Variational Inference

Klaus Greff, Raphaël Lopez Kaufmann, Rishabh Kabra, Nick Watters, Chris Burgess, Daniel Zoran, Loic Matthey, Matthew Botvinick, Alexander Lerchner

slide-2
SLIDE 2

Object Representations Image Representation

Multi-Object Representation Learning

slots VAE IODINE (ours)

slide-3
SLIDE 3

Iterative Object Decomposition Inference NEtwork

  • Built on the VAE framework
  • Incorporates multi-object structure
  • Iterative variational inference

Decoder Structure Iterative Inference

slide-4
SLIDE 4

Iterative Object Decomposition Inference NEtwork

Decoder Structure

  • Built on the VAE framework
  • Incorporates multi-object structure
  • Iterative variational inference

Iterative Inference

slide-5
SLIDE 5

Iterative Object Decomposition Inference NEtwork

Decoder Structure Iterative Inference

  • Built on the VAE framework
  • Incorporates multi-object structure
  • Iterative variational inference
slide-6
SLIDE 6

Object Decomposition by IODINE

I m a g e R e c

  • n

s . M a s k Slot Reconstructions (10) Inpainting of occluded parts

slide-7
SLIDE 7

Multi-Object Disentanglement

VAE IODINE

x y size color material

slide-8
SLIDE 8

Poster #24

Thank you for your Attention!

Summary

  • Multi-object representation learning
  • Completely unsupervised
  • Decomposes an image into individual objects
  • Learns disentangled object representations