Joint image filters are used to transfer structural details from a guidance picture used as a prior to a target image, in tasks such as enhancing spatial resolution and suppressing noise. Previous methods based on convolutional neural networks (CNNs) combine nonlinear activations of spatially-invariant kernels to estimate structural details and regress the filtering result. In this paper, we instead learn explicitly sparse and spatially-variant kernels. We propose a CNN architecture and its efficient implementation, called the deformable kernel network (DKN), that outputs sets of neighbors and the corresponding weights adaptively for each pixel. The filtering result is then computed as a weighted average. We also propose a fast version of DKN that runs about seventeen times faster for an image of size 640 × 480. We demonstrate the effectiveness and flexibility of our models on the tasks of depth map upsampling, saliency map upsampling, cross-modality image restoration, texture removal, and semantic segmentation. In particular, we show that the weighted averaging process with sparsely sampled 3 × 3 kernels outperforms the state of the art by a significant margin in all cases.
Bibliographical noteFunding Information:
The authors would like to thank Yijun Li for helpful discussion. This work was supported in part by Samsung Research Funding & Incubation Center for Future Technology (SRFC-IT1802-06), the Louis Vuitton/ENS chair on artificial intelligence, the Inria/NYU collaboration agreement, and the French government under management of Agence Nationale de la Recherche as part of the “Investissements d’avenir” program, reference ANR-19-P3IA-0001 (PRAIRIE 3IA Institute).
All Science Journal Classification (ASJC) codes
- Computer Vision and Pattern Recognition
- Artificial Intelligence