We propose a novel deep learning architecture for regressing disparity from a rectified pair of stereo images. We leverage knowledge of the problem's geometry to form a cost volume using deep feature representations. We learn to incorporate contextual information using 3-D convolutions over this volume. Disparity values are regressed from the cost volume using a proposed differentiable soft argmin operation, which allows us to train our method end-to-end to sub-pixel accuracy without any additional post-processing or regularization. We evaluate our method on the Scene Flow and KITTI datasets and on KITTI we set a new stateof-the-art benchmark, while being significantly faster than competing approaches.

No abstract

No abstract

The mean low-frequency target strength (TS) of spawning Atlantic herring populations in the Gulf of Maine is estimated from the experimental data acquired during September–October 2006 near the northern flank of Georges Bank. A low-frequency OAWRS system with an instantaneous imaging diameter of 100 km was deployed to provide spatially unaliased imaging of fish populations over wide areas. The OAWRS system’s scattering strength measurements are calibrated with areal fish population density estimates obtained from concurrent localized line-transect measurements with several conventional fish finding sonars (CFFSs). Trawl sampling at selected locations enables the identification of the imaged species. The mean TS estimates of herring individuals exhibits significant variation over OAWRS operating frequency range, in accordance with the results from a resonant scattering model for swimbladder-bearing fish. The neutral buoyancy depth of herring and the species composition in the imaged population is inferred by comparing the measured TS with those derived from the model. Our analysis indicates that the herring population has a neutral buoyancy depth of between 70 and 90 m and is therefore negatively buoyant between 120 and 180 m water depth at which it is commonly found. The herring populations instantaneously imaged with OAWRS often exceeds 200×106, of which over 150×106 individuals can be organized into a large shoal.

No abstract