NeurIPS 2019 Disentanglement Challenge: Improved Disentanglement through Learned Aggregation of Convolutional Feature Maps

2020/02/27 by Maximilian Seitzer, Seitzer, Maximilian, Andreas Foltyn +3
Computer Science · #Computer Vision and Pattern Recognition (cs.CV) #Currency Recognition and Detection #Digital Media Forensic Detection #FOS: Computer and information sciences #Handwritten Text Recognition Techniques #Machine Learning (cs.LG) #Machine Learning (stat.ML)

paper · pdf · doi:10.48550/arxiv.2002.12356

openalex publication_date 2020/02/27 · openalex created_date 2020/03/06 · openalex updated_date 2026/07/28

Abstract

This report to our stage 2 submission to the NeurIPS 2019 disentanglement challenge presents a simple image preprocessing method for learning disentangled latent factors. We propose to train a variational autoencoder on regionally aggregated feature maps obtained from networks pretrained on the ImageNet database, utilizing the implicit inductive bias contained in those features for disentanglement. This bias can be further enhanced by explicitly fine-tuning the feature maps on auxiliary tasks useful for the challenge, such as angle, position estimation, or color classification. Our approach achieved the 2nd place in stage 2 of the challenge. Code is available at https://github.com/mseitzer/neurips2019-disentanglement-challenge.

Citations

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Auto-Encoding Variational Bayes
• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks

Related

• NeurIPS 2019 Disentanglement Challenge: Improved Disentanglement through Aggregated Convolutional Feature Maps
• Unsupervised Disentanglement without Autoencoding: Pitfalls and Future Directions
• Toward a Controllable Disentanglement Network
• Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View
• Disentanglement Challenge: From Regularization to Reconstruction