Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
PairEdit: Learning Semantic Variations for Exemplar-based Image Editing
Authors: Haoguang Lu, Jiacheng Chen, Zhenguo Yang, Aurele Gnanha, Fu Lee Wang, Qing Li, Xudong Mao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive qualitative and quantitative evaluations demonstrate that Pair Edit successfully learns intricate semantics while significantly improving content consistency compared to baseline methods. Code is available at https://github.com/xudonmao/Pair Edit. [...] 4 Experiments [...] 4.2 Results Qualitative Evaluation. Quantitative Evaluation. User Study. [...] 4.3 Ablation Study |
| Researcher Affiliation | Collaboration | Haoguang Lu1 Jiacheng Chen1 Zhenguo Yang2 Aurele Tohokantche Gnanha3 Fu Lee Wang4 Qing Li5 Xudong Mao1 1Sun Yat-sen University 2Guangdong University of Technology 3Huawei Noah s Ark Laboratory 4Hong Kong Metropolitan University 5The Hong Kong Polytechnic University |
| Pseudocode | No | The paper describes the methodology using prose and mathematical equations in Section 3 and its subsections, but it does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps in a code-like format. |
| Open Source Code | Yes | Code is available at https://github.com/xudonmao/Pair Edit. |
| Open Datasets | No | Datasets. We create paired source and target images as follows: First, we apply existing image editing techniques, such as SDEdit [39], to translate source images into preliminary target images. Next, we transfer edited regions from the preliminary target images onto the corresponding regions of source images, generating the final target images. Additionally, some image pairs are collected from the web or sourced from [28]. |
| Dataset Splits | No | For each semantic, we generate 500 pairs of original and edited images using the same random seed across all methods. [...] Our model is trained using either three image pairs (e.g., elf ears, glasses, and chubbiness) or a single image pair (e.g., stylization, dragon eyes, and lipstick). |
| Hardware Specification | Yes | The entire training process takes approximately 8 minutes on a single NVIDIA A100 80GB GPU. |
| Software Dependencies | Yes | Our implementation is based on the publicly available FLUX.1-dev2, with both model weights and text encoders frozen. |
| Experiment Setup | Yes | The rank of Lo RA weights is set to 16. The parameter β is set to 3 for global editing semantics (e.g., stylization) and 1 for local editing semantics (e.g., smile). For all experiments, η and λ are set to 4 and 1, respectively. We jointly train content and semantic Lo RAs for 500 steps using a learning rate of 2 10 3. Following [39, 18], we set the Lo RA scaling factor to 0 during the initial 14 steps to maintain the structure of the original image. |