Limitations of Previous Work

Limitations of Previous Work • Previous deep learning-based image inpainting systems [1, 2, 3, 4] are typically trained on rectangular or unrealistic holes. • They do not support user guidance (e.g., sketch) input for obtaining more user-desired results. E.g., ContextAttention[3] 1. Context encoders: Feature learning by inpainting, Pathak et al., CVPR 2016 2. Globally and Locally Consistent Image Completion, Iizuka et al., TCG 2017 3. Generative Image Inpainting with Contextual Attention, Yu et al., CVPR 2018 4. Image Inpainting for Irregular Holes Using Partial Convolutions, Liu et al., ECCV 2018 ICCV 2018

A presentation slide titled “Limitations of Previous Work” criticizes earlier image-inpainting systems for using rectangular holes and lacking sketch-based user guidance. The examples show faces with square or messy regions removed, making the technical limitation look visibly absurd and funny. It is used as a reaction to awkward presentations, bad software limitations, or hilariously literal demonstrations of a problem. EN: A presentation slide titled “Limitations of Previous Work” criticizes earlier image-inpainting systems for using rectangular holes and lacking sketch-based user guidance. The examples show faces with square or messy regions removed, making the technical limitation look visibly absurd and funny. It is used as a reaction to awkward presentations, bad software limitations, or hilariously literal demonstrations of a problem.

• @ithueti / 1573 2019-10-31