Six-hour CVPR 2020 tutorial is online! Frontier method of new perspective generation

Posted Jun 16, 20202 min read

As one of the world's top three conferences in the field of computer vision, CVPR attracts many professionals to participate every year, and CVPR 2020 is ushering in the grand occasion of over 10,000 submissions. Recently, the CVPR 2020 conference was held online, and nearly 30 tutorials have been launched. Let's join the conference together.

CVPR 2020 was originally scheduled to be held at the conference center in Seattle, Washington from June 16 to 20. However, due to the severe global epidemic situation, the official announcement of CVPR 2020 at the beginning of the year will be held online.

Before the main venue, workshops and teaching courses will be held from June 14-16, this part will be conducted independently by the organizers of each part online.

Nearly thirty teaching courses are already in progress. At present, some courses use YouTube live streaming, and some courses use the ZOOM conference format, some of which have been recorded and shared online.

New perspective generation:from past to future

Today's recommended tutorial"Novel View Synthesis:From Depth-Based Warping to Multi-Plane Images and Beyond"(translation:"New Perspective Synthesis:From Depth-Based Deformation to Multi-Plane Images to the Future") by NVIDIA Host, broadcast live via YouTube, and have uploaded course content.

New perspective synthesis is a long-term challenge in the intersection of computer graphics and computer vision.

The earliest research in this field can be traced back to the 1990s. Early methods of synthesizing new perspectives were:interpolating between corresponding pixels in the input image or between rays in space.

In recent years, deep learning methods have greatly improved the quality of research and brought new research directions to this field.

Tutorial arrangement & organizer

In this tutorial, we first introduced the principles of new perspective generation, including providing background and classification of different methods, and also invited the research team of the latest methods in this field to share.

Easter Eggs:Vernacular Comparison Table

Perspective transformation

Affine transformationaffine transformation


Inverse homography

Depth-based rendering DIBR(depth image-based rendering)

**Single-image-based view generation(SIVG)

Fine network Refinement network

Backward flow map backward flow map

-- Finish --