Learning-based light field reconstruction methods demand in constructing a large receptive field by deepening the network to capture correspondences between input views. In this paper, we propose a spatial-angular attention network to perceive correspondences in the light field non-locally, and reconstruct high angular resolution light field in an end-to-end manner. Motivated by the non-local attention mechanism [1], [2], a spatialangular attention module specifically for the high-dimensional light field data is introduced to compute the responses from all the positions in the epipolar plane for each pixel in the light field, and generate an attention map that captures correspondences along the angular dimension. We then propose a multi-scale reconstruction structure to efficiently implement the non-local attention in the low spatial scale, while also preserving the high frequency components in the high spatial scales. Extensive experiments demonstrate the superior performance of the proposed spatialangular attention network for reconstructing sparsely-sampled light fields with non-Lambertian effects.

Fig 1. We propose a spatial-angular attention module embedded in a multiscale reconstruction structure for learning-based light field reconstruction.

Fig 2. Demonstration of attention map on scenes with (a) large disparity and (b) non-Lambertian effect.

Fig 3. Comparison of the results on the light fields from the CIVIT Dataset (16x upsampling).

Fig 4. Comparison of the results on the light fields from the MPI Light Field Archive (16x upsampling).