In a bustling conference venue, attendees gather around an intriguing poster presentation titled “Wired Perspectives: Multi-View Wire Art Embraces Generative AI.” The poster is part of a display from the University of Surrey, showcasing a project by Zhiyu Qu, Lan Yang, Honggang Zhang, Tao Xiang, Kaiyue Fan, and Yinhai Song. The detailed illustration on the poster explains how their project, DreamWires, integrates AI with multi-view wire art. A diverse audience, including a man in a teal shirt and backpack, engages deeply with the content. Several backpacks and bags rest on the floor nearby, indicating the ongoing exchange of ideas and information. The ceiling above is a network of illuminated grid-like structures, adding to the dynamic atmosphere of intellectual curiosity and exploration.
Text transcribed from the image:
A One ten Diffusion with
Adobe Tan
Richard Dang
UNIVERSITY OF
SURREY
Sketch
Wired Perspectives: Multi-View Wire Art Embraces Generative Al
Zhiyu Qu Lan Yang Honggang Zhang Tao Xiang Kaiyue Pang Yi-Zhe Song
A great thought begins by seeing something differently, with a shift of the mind's eye. Albert Einstein
Overview
Results
DreamWire is the first work that brings Al to Multi-
View Wire Art and makes it infinitely possible.
Isaac Newton
Albert Einstein
Approach
2D
Learnable Bézier
curves parameters
SS...5,
DING
Differentiable
renderer
3D
3D curve
How to render
3D curves?
20 projection
Also accepted by CVPR 2024 Al Art Gallery!
Previous work
La spits
Albert Einstein
DreamWire
Text Control
Val Con
'B'
Hybrid Contr
DreamWire
Matris
3D control points 2D control points
We optimize the 3D Bézier curves over pretrained text-to-image diffusion
models with SDS loss and the following procedures
Calculate the projection of 3D control points and render the projection
curves through a differentiable 2D Bézier curve renderer
Input all the rendered sketches from different perspectives into T25
diffusion models with different text prompts
(Optional) The spatial continuity of curves is assured by a designed
Minimum Spanning Tree (MST) loss.
The effect of M
EL51
D