EditGAN: High-Precision Semantic Image Editing

Authors: Huan Ling, Karsten Kreis, Daiqing Li, Seung Wook Kim, Antonio Torralba, Sanja Fidler

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally show that Edit GAN can manipulate images with an unprecedented level of detail and freedom, while preserving full image quality.We can also easily combine multiple edits and perform plausible edits beyond Edit GAN s training data. We demonstrate Edit GAN on a wide variety of image types and quantitatively outperform several previous editing methods on standard editing benchmark tasks.
Researcher Affiliation Collaboration Huan Ling1,2,3, Karsten Kreis1, Daiqing Li 1 Seung Wook Kim1,2,3 Antonio Torralba4 Sanja Fidler1,2,3 1NVIDIA 2University of Toronto 3Vector Institute 4MIT {huling,kkreis,daiqingl,seungwookk,sfidler}@nvidia.com, torralba@mit.edu
Pseudocode No The paper describes its methods and processes using textual descriptions and diagrams (e.g., Figure 3), but does not contain a formally structured pseudocode block or algorithm section.
Open Source Code No The paper mentions 'Project page: https://nv-tlabs.github.io/edit GAN.' which is a project overview page, not a direct link to a source-code repository for the methodology.
Open Datasets Yes We extensively evaluate Edit GAN on images across four different categories: Cars (384 × 512 spatial resolution), Birds (512 × 512), Cats (256 × 256), and Faces (1024 × 1024). ... For Car, Cat, and Faces, we use real images from Dataset GAN s test set that were not part of GAN training to demonstrate editing functionality. These images are first embedded into Edit GAN s latent space via an encoder and optimization as described in Sec. 3.2. For Birds, we show editing on GAN-generated images. Model details and hyperparameters are provided in the Appendix.
Dataset Splits No The paper mentions 'We train our segmentation branch as described in Sec. 3.2 using 16, 16, 30, and 30 image-mask pairs as labeled training data for Faces, Cars, Birds, and Cats, respectively.' and refers to a 'test set' from Dataset GAN, but does not specify validation splits or percentages for data used in its own experiments.
Hardware Specification Yes We carefully measure the run time of our editing on an NVIDIA Tesla V100 GPU.
Software Dependencies Yes pytorch-fid: FID Score for Py Torch. https://github.com/mseitzer/ pytorch-fid, August 2020. Version 0.1.1.
Experiment Setup Yes When editing is done purely optimization-based or when learning the editing vectors, we always perform 100 steps of optimization using Adam [77].