FastDrag: Manipulate Anything in One Step

Authors: Xuanjia Zhao, Jian Guan, Congyi Fan, Dongli Xu, Youtian Lin, Haiwei Pan, Pengming Feng

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our Fast Drag is validated on the Drag Bench dataset, demonstrating substantial improvements in processing time over existing methods, while achieving enhanced editing performance.
Researcher Affiliation Academia 1College of Computer Science and Technology, Harbin Engineering University 2School of Intelligence Science and Technology, Nanjing University 3State Key Laboratory of Space-Ground Integrated Information Technology 4Independent Researcher
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Project page: https://fastdrag-site.github.io/. We include the implementation code for our method in the supplementary materials submitted.
Open Datasets Yes We conduct quantitative comparison using Drag Bench dataset [28], which consists of 205 different types of images with 349 pairs of handle and target points.
Dataset Splits No The paper uses the Drag Bench dataset for quantitative comparison and ablation studies but does not specify any training, validation, or test splits for its experiments.
Hardware Specification Yes Our experiments are conducted on an RTX 3090 GPU with 24G memory.
Software Dependencies No The paper mentions 'Stable Diffusion 1.5' but does not provide specific version numbers for programming languages or libraries like Python, PyTorch, or CUDA.
Experiment Setup Yes Unless otherwise specified, the default setting for inversion and sampling step is 10. Following Drag Diffusion [28], classifier-free guidance (CFG) [8] is not applied in diffusion model, and we optimize the diffusion latent at the 7th step.