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..
UltraHR-100K: Enhancing UHR Image Synthesis with A Large-Scale High-Quality Dataset
Authors: Chen Zhao, En Ci, Yunzhe Xu, Tiehan Fan, Shanyan Guan, Yanhao Ge, Jian Yang, Ying Tai
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on our proposed Ultra HR-eval4K benchmarks demonstrate that our approach significantly improves the fine-grained detail quality and overall fidelity of UHR image generation. |
| Researcher Affiliation | Collaboration | 1 State Key Laboratory of Novel Software Technology, Nanjing University, China 2 vivo Mobile Communication Co., Ltd., China |
| Pseudocode | No | The paper describes methods like Detail-Oriented Timestep Sampling (DOTS) and Soft-Weighting Frequency Regularization (SWFR) textually and mathematically in Section 4, but it does not present them in structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at here. |
| Open Datasets | No | The paper introduces a new dataset, Ultra HR-100K, and a benchmark, Ultra HR-eval4K, but does not provide concrete access information such as a specific URL, DOI, repository name, or explicit mention of its availability in supplementary materials for the dataset itself. The code is stated to be available in supplementary material, but not the dataset. |
| Dataset Splits | Yes | In addition, we construct a evaluation subset from our dataset Ultra HR-eval4K containing 2,000 images. |
| Hardware Specification | Yes | Due to computational constraints, we conduct training solely on SANA, and all experiments are performed on four H20 GPUs. |
| Software Dependencies | No | The paper mentions using the CAMEWrapper [15] optimizer, but does not specify its version or any other software dependencies with version numbers. |
| Experiment Setup | Yes | We use the CAMEWrapper [15] optimizer with a constant learning rate of 1e-4, and employ mixed-precision training with a batch size of 24. The first-stage training is conducted for 4K iterations, followed by 8K iterations in the second stage. |