Upsampling

About: Upsampling is a research topic. Over the lifetime, 2426 publications have been published within this topic receiving 57613 citations.

Embracing Single Stride 3D Object Detector with Sparse Transformer

Abstract: In LiDAR-based 3D object detection for autonomous driving, the ratio of the object size to input scene size is significantly smaller compared to 2D detection cases. Over-looking this difference, many 3D detectors directly follow the common practice of 2D detectors, which downsample the feature maps even after quantizing the point clouds. In this paper, we start by rethinking how such multi-stride stereotype affects the LiDAR-based 3D object detectors. Our experiments point out that the downsampling operations bring few advantages, and lead to inevitable information loss. To remedy this issue, we propose Single-stride Sparse Transformer (SST) to maintain the original resolution from the beginning to the end of the network. Armed with transformers, our method addresses the problem of insufficient receptive field in single-stride architectures. It also cooperates well with the sparsity of point clouds and naturally avoids expensive computation. Eventually, our SST achieves state-of-the-art results on the large-scale Waymo Open Dataset. It is worth mentioning that our method can achieve exciting performance (83.8 LEVEL_1 AP on validation split) on small object (pedestrian) detection due to the characteristic of single stride. Our codes will be public soon.

Sub-band speech coding system

Abstract: An improved sub-band speech coding system is provided by subdividing signals into a lower an higher subband, downsampling the lower subband before coding and coding the higher subband without downsampling. The decoder includes decoding and upsampling of the lower subband and decoding the higher subband and adding the higher subband to the lower subband.

Directional Super-Resolution by Means of Coded Sampling and Guided Upsampling

Abstract: We present a simple guided super-resolution technique for increasing directional resolution without reliance on depth estimation or image correspondences. Rather, it searches for best- matching multidimensional (4D or 3D) patches within the entire captured data set to compose new directional images that are consistent in both the spatial and the directional domains. We describe algorithms for guided upsampling, iterative guided upsampling, and sampling code estimation. Our experimental results reveal that the outcomes of existing light-field camera arrays and light-stage systems can be improved without additional hardware requirements or recording effort simply by realignment of cameras or light sources to change their sampling patterns.

Interpolating filter banks in arbitrary dimensions

Abstract: Interpolating filter banks are constructed for use with signals which may be represented as a lattice of arbitrary dimension d. The filter banks include M channels, where M is greater than or equal to two. A given filter bank is built by first computing a set of shifts τ i as D -1 t i , i=1,2, . . . M-1, where t i is a set of coset representatives taken from a unit cell of the input signal lattice, and D is a dilation matrix having a determinant equal to M. A polynomial interpolation algorithm is then applied to determine weights for a set of M-1 predict filters P i having the shifts τ i . A corresponding set of update filters U i are then selected as U i =P i */M, where P i * is the adjoint of the predict filter P i . The resulting predict and update filters are arranged in a lifting structure such that each of the predict and update filters are associated with a pair of the M channels of the filter bank. The input signal applied to the filter bank is downsampled in each of the M channels, and then interpolated using the M-1 predict filters and the M-1 update filters. The downsampled and interpolated signal may be reconstructed using complementary interpolation and upsampling operations.