Compensated isocurvature perturbations (CIPs) are primordial fluctuations that balance baryon and dark-matter isocurvature to leave the total matter density unperturbed. The effects of CIPs on the cosmic microwave background (CMB) anisotropies are similar to those produced by weak lensing of the CMB: smoothing of the power spectrum, and generation of non-Gaussian features. Previous work considered the CIP effects on the CMB power-spectrum but neglected to include the CIP effects on estimates of the lensing potential power spectrum (though its contribution to the non-Gaussian, connected, part of the CMB trispectrum). Here, the CIP contribution to the standard estimator for the lensing potential power-spectrum is derived, and along with the CIP contributions to the CMB power-spectrum, Planck data is used to place limits on the root-mean-square CIP fluctuations on CMB scales, ∆ 2 rms (RCMB). The resulting constraint of ∆ 2 rms (RCMB) < 4.3 × 10 −3 using this new technique improves on past work by a factor of ∼ 3. We find that for Planck data our constraints almost reach the sensitivity of the optimal CIP estimator. The method presented here is currently the most sensitive probe of the amplitude of a scale-invariant CIP power spectrum placing an upper limit of ACIP < 0.017 at 95% CL. Future measurements of the large-scale CMB lensing potential power spectrum could probe CIP amplitudes as low as ∆ 2 rms (RCMB) = 8 × 10 −5 (ACIP = 3.2 × 10 −4 ).

Modern neural network-based algorithms are able to produce highly accurate depth estimates from stereo image pairs, nearly matching the reliability of measurements from more expensive depth sensors. However, this accuracy comes with a higher computational cost since these methods use network architectures designed to compute and process matching scores across all candidate matches at all locations, with floating point computations repeated across a match volume with dimensions corresponding to both space and disparity. This leads to longer running times to process each image pair, making them impractical for real-time use in robots and autonomous vehicles. We propose a new stereo algorithm that employs a significantly more efficient network architecture. Our method builds an initial match cost volume using traditional matching costs that are fast to compute, and trains a network to estimate disparity from this volume. Crucially, our network only employs per-pixel and two-dimensional convolution operations: to summarize the match information at each location as a low-dimensional feature vector, and to spatially process these "cost-signature" features to produce a dense disparity map. Experimental results on the KITTI benchmark show that our method delivers competitive accuracy at significantly higher speeds-running at 48 frames per second on a modern GPU.