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..
Text to Sketch Generation with Multi-Styles
Authors: Tengjie Li, Shikui Tu, Lei Xu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that our approach achieves high-quality sketch generation with accurate style alignment and improved flexibility in style control. The official implementation of M3S is available at https://github.com/CMACH508/M3S. ... Table 1 reports the quantitative results of different methods. |
| Researcher Affiliation | Academia | Tengjie Li 1, Shikui Tu1 , Lei Xu1,2 1School of Computer Science, Shanghai Jiao Tong University 2 Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ), Guangdong, China EMAIL |
| Pseudocode | No | The paper describes its methodology using equations and textual descriptions, but it does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | The official implementation of M3S is available at https://github.com/CMACH508/M3S. |
| Open Datasets | Yes | We evaluate different methods for single-style referenced generation on six diverse sketch datasets encompassing professional, amateur, and abstract styles: four professional styles from 4skst [37] ... and 50 abstract freehand sketches from Sketchy [35]. ... We conducted an additional experiment set pairing one randomly selected S5 image with one randomly chosen image from the Quick Draw (QD) dataset [11] per prompt. |
| Dataset Splits | No | The paper describes a training-free framework and how it selects reference sketches and prompts for evaluation, but it does not specify traditional training, validation, or test dataset splits for a model. For systematic testing, we generate 50 textual prompts via Deep Seek [27] using the template "A sketch of ...", pairing each prompt with a randomly selected referenced sketch. |
| Hardware Specification | Yes | Each sketch takes about 40 seconds (M3S (SD v1.5)) and 70 seconds (M3S (SDXL)) on an A100 40GB GPU. |
| Software Dependencies | No | The paper states: "We implement M3S on both Stable Diffusion v1.5 [33] and SDXL [29]", but does not provide specific version numbers for ancillary software like programming languages, libraries (e.g., PyTorch), or other dependencies. |
| Experiment Setup | Yes | For M3S (SD v1.5), we configure ω1 = 15, ω2 = 15 and λ = 0.1 Styles 1-5, while using ω1 = 15, ω2 = 25 and λ = 0.05 for Style 6. ... For M3S (SDXL), we set ω1 = 15, ω2 = 15 and λ = 0.1 for Styles 1-5, adjusting to ω1 = 7.5, ω2 = 20 and λ = 0.05 for Style 6. ... We applied DDIM [40] to sample the target sketches with 100 steps. In practice, we empirically set γ = 60 as the default and clamp gradx and grady within [ 0.001, 0.001]... |