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..
Linearly Constrained Diffusion Implicit Models
Authors: Vivek Jayaram, Ira Kemelmacher-Shlizerman, Steve Seitz, John Thickstun
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments to demonstrate the efficiency and quality of CDIM across various tasks and datasets. In Section 5.1, we present quantitative comparisons to state-of-the-art approaches, followed by a comparison against DPS using DDIM in Section 5.2. In Section 5.3 we describe ablation studies examining inference speed and hyperparameters. Finally, in Section 5.4 we explore two novel applications of diffusion models for inverse problems. |
| Researcher Affiliation | Academia | 1University of Washington 2Cornell University |
| Pseudocode | Yes | Algorithm 1 Constrained Diffusion Implicit Models |
| Open Source Code | Yes | Code and an interactive demo can be found on our project website. 1 1https://grail.cs.washington.edu/projects/cdim/ |
| Open Datasets | Yes | We evaluate CDIM on the FFHQ-1k [35] and Image Net-1k [36] validation sets. |
| Dataset Splits | Yes | We evaluate CDIM on the FFHQ-1k [35] and Image Net-1k [36] validation sets. |
| Hardware Specification | Yes | All experiments are carried out on a single Nvidia A100 GPU. |
| Software Dependencies | No | The paper mentions specific diffusion models and frameworks (e.g., DDPM [1], DDIM [7], DPS [2]), but does not list specific versions of software dependencies like programming languages, libraries, or operating systems used for their implementation. |
| Experiment Setup | Yes | We present results using our method with both T = 25 denoising steps and T = 50 denoising steps. In all cases we use c = 0.1 for the number of standard deviations in the stopping criteria. For all tasks, we apply zero-centered Gaussian measurement noise with σ = 0.05. CDIM only contains two hyperparameters: the number of denoising steps T and the plausible region stopping criteria constant c multiplied by σt(y). |