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..
Uni-Instruct: One-step Diffusion Model through Unified Diffusion Divergence Instruction
Authors: Yifei Wang, Weimin Bai, colin zhang, Debing Zhang, Weijian Luo, He Sun
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | On the CIFAR10 generation benchmark, Uni-Instruct achieves record-breaking Frechet Inception Distance (FID) values of 1.46 for unconditional generation and 1.38 for conditional generation. On the Image Net 64 64 generation benchmark, Uni-Instruct achieves a new So TA one-step generation FID of 1.02, which outperforms its 79-step teacher diffusion with a significant improvement margin of 1.33 (1.02 vs 2.35). We also apply Uni-Instruct on broader tasks like text-to-3D generation, which slightly outperform previous methods, such as SDS and VSD, in terms of both generation quality and diversity. Both the solid theoretical and empirical contributions of Uni-Instruct will potentially help future studies on onestep diffusion distillation and knowledge transfer of diffusion models. |
| Researcher Affiliation | Collaboration | Yifei Wang1,5 Weimin Bai1,2,3 Colin Zhang4 Debing Zhang4 Weijian Luo4 He Sun1,2,3 1College of Future Technology, Peking University 2National Biomedical Imaging Center, Peking University 3Academy for Advanced Interdisciplinary Studies, Peking University 4 hi-lab, Xiaohongshu Inc 5 Yuanpei College, Peking University |
| Pseudocode | Yes | Algorithm 1: Uni-Instruct Algorithm on Distilling One Step Diffusion Model Algorithm 2: Uni-Instruct for Text-to-3D Generation |
| Open Source Code | No | Code will be available at Github. |
| Open Datasets | Yes | Experiment Settings We evaluate Uni-Instruct for both conditional and unconditional generations on CIFAR10 [22] and conditional generations on Image Net 64 64[8]. [8] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A largescale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248 255. Ieee, 2009. [22] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. |
| Dataset Splits | No | The paper uses standard benchmark datasets (CIFAR10, Image Net 64 64) but does not explicitly state the training/validation/test splits within the text. |
| Hardware Specification | No | The paper mentions that computational resources were discussed in the appendix, but the appendix does not provide specific hardware details (e.g., GPU models, CPU types) used for the experiments. |
| Software Dependencies | No | The paper mentions using EDM [19] as teacher models and Adam [21] as an optimizer, but does not provide specific version numbers for these or other software libraries (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | Experiment Settings We evaluate Uni-Instruct for both conditional and unconditional generations on CIFAR10 [22] and conditional generations on Image Net 64 64[8]. We use EDM [19] as teacher models. In each experiment, we implement three types of divergences: Reverse-KL (RKL), Forward KL (FKL), and Jeffrey-KL (JKL) divergence. We borrow the parameters settings from Si DA [68], which takes the output from the diffusion unet encoder directly as the discriminator. As for evaluation metrics, we use FID, as it simultaneously quantifies both image quality and diversity. Algorithm 2: Uni-Instruct for Text-to-3D Generation ... train the Ne RF model for 300 400 epochs ... fine-tune the object s geometry appearance for 150 epochs ... further finetuning with Uni-Instruct guidance for an additional 150 epochs. |