Specifying Object Attributes and Relations in Interactive Scene Generation
j20232 opened this issue · 0 comments
Summary
- Image generation tool in which the input consists of a scene graph with the potential addition of location information
- Each object is associated both with a location embedding and with an appearance embedding
- The latter can be extracted from another image, allowing for a duplication of existing objects to a new image, where their layout is drastically changed
- In addition to the dual encoding, our method presents both a new architecture and new loss terms, which leads to an improved performance over the existing baselines
Abstract
We introduce a method for the generation of images from an input scene graph. The method separates between a layout embedding and an appearance embedding. The dual embedding leads to generated images that better match the scene graph, have higher visual quality, and support more complex scene graphs. In addition, the embedding scheme supports multiple and diverse output images per scene graph, which can be further controlled by the user. We demonstrate two modes of per-object control: (i) importing elements from other images, and (ii) navigation in the object space, by selecting an appearance archetype. Our code is publicly available at this https URL
Author
Journal/Conference
The IEEE International Conference on Computer Vision (ICCV), 2019
Subjects
cs.CV: Computer Vision and Pattern Recognitioncs.LG: Machine Learning
Comment
Best Paper Honorable Mention in ICCV 2019