Meta tags:

Headings (most frequently used words):

volumetric, correspondence, networks, for, optical, flow, bibtex, acknowledgments, abstract,

Text of the page (most frequently used words):
for (10), and (9), the (8), volumetric (8), networks (8), flow (6), #correspondence (5), #optical (5), that (5), use (4), click (3), high (3), res (3), video (3), here (3), test (3), such (3), stereo (3), search (3), sota (3), layers (3), accuracy (3), work (2), argo (2), yang (2), gengshan (2), ramanan (2), deva (2), processing (2), 2019 (2), bibtex (2), tasks (2), can (2), make (2), cost (2), window (2), also (2), dramatically (2), capture (2), multi (2), significantly (2), while (2), our (2), this, was, supported, cmu, center, autonomous, vehicle, research, acknowledgments, inproceedings, yang2019volumetric, title, author, booktitle, advances, neural, information, systems, pages, 793, 803, year, tum, plant, kitti, 140, sintel, clean, ambush, poster, slides, code, paper, many, classic, vision, estimation, disparities, cast, dense, matching, well, known, techniques, doing, volume, typically, tensor, match, costs, between, all, pixels, image, their, potential, matches, state, art, deep, representations, internal, however, require, significant, amounts, memory, compute, making, them, cumbersome, practice, result, employ, various, heuristics, designed, limit, leading, limited, overfitting, instead, introduce, several, simple, modifications, simplify, encoder, decoder, architectures, efficiently, large, receptive, fields, channel, volumes, dimensional, notions, pixel, similarities, finally, separable, filtering, reduces, computation, parameters, preserving, innovations, improve, over, standard, benchmarks, being, easier, with, training, converges, fewer, iterations, most, importantly, generalize, across, fly, adaptation, windows, allows, repurpose, vice, versa, used, implement, adaptive, increase, sizes, demand, abstract, robotics, institute, carnegie, mellon, university, neurips,

Text of the page (random words):
volumetric correspondence networks for optical flow volumetric correspondence networks for optical flow neurips 2019 gengshan yang 1 deva ramanan 1 2 1 robotics institute carnegie mellon university 2 argo ai abstract many classic tasks in vision such as the estimation of optical flow or stereo disparities can be cast as dense correspondence matching well known techniques for doing so make use of a cost volume typically a 4d tensor of match costs between all pixels in a 2d image and their potential matches in a 2d search window state of the art sota deep networks for flow stereo make use of such volumetric representations as internal layers however such layers require significant amounts of memory and compute making them cumbersome to use in practice as a result sota networks also employ various heuristics designed to limit volumetric processing leading to limited accuracy and overfitting instead we introduce several simple modifications that dramatically simplify the use of volumetric layers 1 volumetric encoder decoder architectures that efficiently capture large receptive fields 2 multi channel cost volumes that capture multi dimensional notions of pixel similarities and finally 3 separable volumetric filtering that significantly reduces computation and parameters while preserving accuracy our innovations dramatically improve accuracy over sota on standard benchmarks while being significantly easier to work with training converges in 7x fewer iterations and most importantly our networks generalize across correspondence tasks on the fly adaptation of search windows allows us to repurpose optical flow networks for stereo and vice versa and can also be used to implement adaptive networks that increase search window sizes on demand paper code slides poster bibtex sintel clean ambush 3 test click here for the high res video kitti 140 test click here for the high res video tum plant test click here for the high res video bibtex inproceedings yang2019volumetric title volumetric correspondence networks for optical flow author yang gengshan and ramanan deva booktitle advances in neural information processing systems pages 793 803 year 2019 acknowledgments this work was supported by the cmu argo ai center for autonomous vehicle research