Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Training an optical flow network using end-to-end learning

IP.com Disclosure Number: IPCOM000248288D
Publication Date: 2016-Nov-15
Document File: 2 page(s) / 32K

Publishing Venue

The IP.com Prior Art Database

Abstract

Optical flow estimation is a building-block in computer vision and visual analytics. We present a method for learning an optical flow estimatior by using video data or a collection of images without annotating thier correspondence relationship. The method enables an end-to-end training of a neural network which is used once trained for optical flow estimation.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 100% of the total text.

Page 01 of 2

Training an optical flow network using end-to-end learning

The network consists of several important components as follows:

 The key points are:
1) Training an optical flow network model with unlabeled image data, which involves the following key components and steps


•Flow offset regression network

1


Page 02 of 2

  •Interpolation network •Loss function with a data and smooth term using the warped image and the target image Training the above optical flow network using a pair of relevant images (without label) based on differential back propagation.

2