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..
DISCO: DISCrete nOise for Conditional Control in Text-to-Image Diffusion Models
Authors: Longquan Dai, Wu Ming, Dejiao Xue, wang he, Jinhui Tang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results demonstrate that DISCO achieves competitive controllability while substantially reducing resource demands, positioning it as a scalable and effective alternative for conditional diffusion generation. This section presents a comprehensive quantitative and qualitative evaluation of our method, emphasizing its training efficiency and enhanced alignment performance. |
| Researcher Affiliation | Academia | Longquan Dai, Wu Ming, Dejiao Xue, He Wang, and Jinhui Tang Nanjing University of Science and Technology, Nanjing, China EMAIL |
| Pseudocode | No | The paper describes the methodology in prose, but it does not contain a clearly labeled pseudocode or algorithm block, nor structured steps formatted like code. |
| Open Source Code | Yes | Code is available at https://github.com/dailongquan/disco. |
| Open Datasets | Yes | We utilize approximately 118,000 images from the COCO2017 dataset [19] across all conditioning types. |
| Dataset Splits | No | The paper mentions using the COCO2017 dataset and a subset for human pose estimation. However, it does not explicitly provide specific training/test/validation split percentages, sample counts, or clear references to how the COCO dataset was partitioned for experiments. |
| Hardware Specification | Yes | Our training process completes in 3 days on a single NVIDIA RTX 3090 (24GB). |
| Software Dependencies | No | The paper mentions 'Adam optimizer' but does not specify its version. No other software components are listed with specific version numbers. |
| Experiment Setup | Yes | We train the discrete noise prediction network for 35 epochs with a batch size of 32, using the Adam optimizer with a learning rate of 1 10 5. During training, both input images and condition maps are resized to 512 512. |