ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing

Authors: Bingchuan Li, Tianxiang Ma, Peng Zhang, Miao Hua, Wei Liu, Qian He, Zili Yi

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that our approach yields near-perfect reconstructions without sacrificing the editability, thus allowing accurate manipulation of real images. Further, we evaluate the performance of our rectifying network, and see great generalizability towards unseen manipulation types and out-of-domain images.
Researcher Affiliation Industry Bingchuan Li, Tianxiang Ma, Peng Zhang, Miao Hua, Wei Liu, Qian He, Zili Yi* Byte Dance Ltd, Beijing, China {libingchuan, matianxiang.724, liuwei.jikun, zhangpeng.ucas, heqian, huamiao, yizili}@bytedance.com
Pseudocode Yes Algorithm 1: Training & inference
Open Source Code No The paper does not contain an explicit statement about releasing the source code or provide a link to a code repository for the methodology described.
Open Datasets Yes To demonstrate the generalizability of our method on unseen realistic faces, we train our model on the FFHQ (Karras, Laine, and Aila 2019) dataset, and test on the Celeb A-HQ (Karras et al. 2018) dataset. We also experiment our method on the animal portrait dataset (Choi et al. 2020) and the anime dataset (Branwen 2019).
Dataset Splits No The paper mentions training on FFHQ and testing on Celeb A-HQ but does not provide specific train/validation/test dataset split percentages or sample counts.
Hardware Specification Yes The proposed Re GANIE network is trained on a single Tesla V100 GPU with batch size of 8 on 512 512 resolution images.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies, such as programming languages, libraries, or frameworks.
Experiment Setup Yes The proposed Re GANIE network is trained on a single Tesla V100 GPU with batch size of 8 on 512 512 resolution images. It is optimized using the Adam optimizer (Kingma and Ba 2014) with b1 = 0.0 and b2 = 0.99, and the learning rate is fixed at 0.002. In all experiments, the training is performed for 100,000 iterations.