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..
DCI: Dual-Conditional Inversion for Boosting Diffusion-Based Image Editing
Authors: Zixiang Li, Haoyu Wang, Wei Wang, Chuangchuang Tan, Yunchao Wei, Yao Zhao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments to evaluate the effectiveness of Dual-Conditional Inversion (DCI). This section is organized as follows. In Section 4.1, we introduce the datasets, evaluation metrics and experimental settings. Section 4.2 compares DCI with representative inversion methods across multiple aspects quantitatively and qualitatively. In Section 4.3, we investigate how DCI reduces both the latent noise gap and the reconstruction error. Finally, Section 4.4 presents an ablation study to assess the impact of key hyperparameters and design choices. |
| Researcher Affiliation | Academia | Zixiang Li1,2, Haoyu Wang1,2, Wei Wang1,2, Chuangchuang Tan1,2, Yunchao Wei1,2, Yao Zhao1,2 1Institute of Information Science, Beijing Jiaotong University 2Visual Intelligence +X International Cooperation Joint Laboratory of MOE |
| Pseudocode | Yes | Algorithm 1 Dual-Conditional Inversion (DCI) Input: Source image latent z0, DDIM steps T, source prompt ps, maximal optimization rounds K, threshold δ, image guidance strength λ, fixed-point learning rate η, reference noise ϵref Output: Inversion noise z T 1: for t = 1 to T do 2: for i = 1 to K do 3: Get zt from zt 1 based on (3) 4: Predict noise ˆϵraw based on (4) 5: Compute Lref = ˆϵraw ϵref 2 6: Apply correction: ˆϵ = ˆϵraw λ ˆϵraw Lref 7: Update zt using ˆϵ 8: Calculate Lfix = fθ(zt) zt 2 9: Update zt = zt η zt Lfix 10: if Lfix < δ then break end if 11: end for 12: end for |
| Open Source Code | Yes | Our codes are available at: https://github.com/Lzxhh/Dual-Conditional-Inversion |
| Open Datasets | Yes | We verifies the effectiveness of our proposed DCI method mainly on the PIE-Bench [20], which comprises 700 images featuring 10 distinct editing types. ... We also use the COCO2017 [28] to test the application of our method in a wider range of scenarios. |
| Dataset Splits | No | The paper mentions using PIE-Bench and COCO2017 for evaluation and randomly selecting 100 captions from PIE-Bench for specific analysis. However, it does not provide explicit training/test/validation splits for the main experiments, nor does it specify how the datasets are partitioned for reproducing the results, beyond using the full datasets for evaluation. |
| Hardware Specification | Yes | All experiments and validations are conducted on a single NVIDIA RTX 4090 GPU. |
| Software Dependencies | No | The paper mentions using Stable Diffusion v1.4 as the base model, but does not provide specific version numbers for other ancillary software components like programming languages, libraries (e.g., PyTorch, TensorFlow), or CUDA versions. |
| Experiment Setup | Yes | In our experiments, we utilize Stable Diffusion v1.4 as the base model with DDIM sampling steps of 50 and a Classifier-Free Guidance (CFG) scale of 7.5. These settings are the same as those used in the baselines. For DCI, we set the hyper-parameters to K = 5, λ = 2, and η = 0.001. |