Poster
Dynamic Filter Networks
Xu Jia · Bert De Brabandere · Tinne Tuytelaars · Luc V Gool

Mon Dec 5th 06:00 -- 09:30 PM @ Area 5+6+7+8 #71 #None
in Monday Posters »

In a traditional convolutional layer, the learned filters stay fixed after training. In contrast, we introduce a new framework, the Dynamic Filter Network, where filters are generated dynamically conditioned on an input. We show that this architecture is a powerful one, with increased flexibility thanks to its adaptive nature, yet without an excessive increase in the number of model parameters. A wide variety of filtering operation can be learned this way, including local spatial transformations, but also others like selective (de)blurring or adaptive feature extraction. Moreover, multiple such layers can be combined, e.g. in a recurrent architecture. We demonstrate the effectiveness of the dynamic filter network on the tasks of video and stereo prediction, and reach state-of-the-art performance on the moving MNIST dataset with a much smaller model. By visualizing the learned filters, we illustrate that the network has picked up flow information by only looking at unlabelled training data. This suggests that the network can be used to pretrain networks for various supervised tasks in an unsupervised way, like optical flow and depth estimation.

Keywords: Deep Learning or Neural Networks · (Application) Computer Vision · (Other) Unsupervised Learning Methods

Author Information

Xu Jia (KU Leuven)
Bert De Brabandere (KU Leuven)
Tinne Tuytelaars (KU Leuven)
Luc V Gool (ETH Zürich)

More from the Same Authors