Permutohedral Lattice CNNs
This work addresses a domain-specific challenge in image recognition by generalizing convolution for sparse features, though it appears incremental as it builds on existing bilateral filter techniques.
The paper tackles the problem of processing sparse input features in convolutional networks by introducing a convolutional layer that uses the permutohedral lattice data structure, enabling efficient filtering of non-grid signals like color values.
This paper presents a convolutional layer that is able to process sparse input features. As an example, for image recognition problems this allows an efficient filtering of signals that do not lie on a dense grid (like pixel position), but of more general features (such as color values). The presented algorithm makes use of the permutohedral lattice data structure. The permutohedral lattice was introduced to efficiently implement a bilateral filter, a commonly used image processing operation. Its use allows for a generalization of the convolution type found in current (spatial) convolutional network architectures.