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 [1].
Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation
Authors: Cheng Jin, Zhenyu Xiao, Chutao Liu, Yuantao Gu
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that ADG significantly outperforms existing methods, generating images that not only maintain superior text alignment but also exhibit improved color fidelity and better alignment with human perceptual preferences. |
| Researcher Affiliation | Academia | 1Department of Electronic Engineering, Tsinghua University, Beijing, China 2Zhili College, Tsinghua University, Beijing, China. Correspondence to: Yuantao Gu <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Angle-Domain Guidance Sampling (ADG) |
| Open Source Code | Yes | The implementation is available at github.com/jinc7461/ADG. |
| Open Datasets | Yes | Experimental results on the COCO dataset demonstrate that the ADG algorithm outperforms the baseline, particularly in generating images that are better aligned with textual descriptions, underscoring its potential to overcome the limitations of existing methods. |
| Dataset Splits | No | The paper uses the COCO dataset, specifically referencing 'COCO10k', but does not explicitly detail the training/validation/test splits used for the experiments (e.g., percentages or sample counts). |
| Hardware Specification | No | No specific hardware details such as GPU models, CPU types, or memory specifications are mentioned for the experimental setup. |
| Software Dependencies | No | The paper mentions Stable Diffusion v3.5 and DPM-Solver sampler but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | Results of 10 NFE generation with SD v3.5 (d=38) on COCO10k |