OCONet: Image Extrapolation by Object Completion
Authors:
- 1 June 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2307-2317
Abstract
Image extrapolation extends an input image beyond the originally-captured field of view. Existing methods struggle to extrapolate images with salient objects in the foreground or are limited to very specific objects such as humans, but tend to work well on indoor/outdoor scenes. We introduce OCONet (Object COmpletion Networks) to extrapolate foreground objects, with an object completion network conditioned on its class. OCONet uses an encoder-decoder architecture trained with adversarial loss to predict the object’s texture as well as its extent, represented as a predicted signed-distance field. An independent step extends the background, and the object is composited on top using the predicted mask. Both qualitative and quantitative results show that we improve on state-of-the-art image extrapolation results for challenging examples.Keywords
This publication has 31 references indexed in Scilit:
- BiggerPictureACM Transactions on Graphics, 2014
- Photo Uncrop2014
- FrameBreak: Dramatic Image Extrapolation by Guided Shift-Maps2013
- PatchMatchACM Transactions on Graphics, 2009
- Seam carving for content-aware image resizingACM Transactions on Graphics, 2007
- Image Alignment and Stitching: A TutorialFoundations and Trends in Computer Graphics and Vision, 2007
- Automatic Panoramic Image Stitching using Invariant FeaturesInternational Journal of Computer Vision, 2006
- Interactive digital photomontage2004
- Poisson image editingACM Transactions on Graphics, 2003
- Shape modeling with front propagation: a level set approachIEEE Transactions on Pattern Analysis and Machine Intelligence, 1995