Dynamic Filter Networks

TL;DR

This paper introduces Dynamic Filter Networks, which generate adaptive filters conditioned on input data, enhancing flexibility and efficiency in tasks like video prediction and optical flow estimation.

Contribution

The paper presents a novel framework for dynamically generating filters conditioned on input, enabling flexible filtering operations without significantly increasing model complexity.

Findings

Achieved state-of-the-art results on moving MNIST with a smaller model

Demonstrated effectiveness in video and stereo prediction tasks

Visualized learned filters to show flow information capture

Abstract

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 operations 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.