Skip to main content

Participation at EICS 2025

· 3 min read
Tommaso Calò
Tommaso Calò
Ph.D. Student

From June 23 to June 27, 2025, the e-Lite group will attend the 17th ACM SIGCHI Symposium on Engineering Interactive Computing Systems (EICS 2025) in Trier, Germany.

Francesca and Tommaso will attend the conference in person to showcase their latest work on enhancing the workflow for UI designers and developers using AI. Tommaso will present a paper introducing new datasets to train more capable AI models that can automatically generate code from a visual concept. Francesca will present a late-breaking result that introduces a method to translate web data into Figma-compatible designs, allowing the development of AI-powered assistants that integrate directly into designers' creative workflows.

If you see any of us, feel free to say hello: we're always happy to chat! 😊

EICS 2025 conference banner - premier international forum for engineering interactive computing systems

Advancing Code Generation from Visual Designs through Transformer-Based Architectures and Specialized Datasets​

The paper "Advancing Code Generation from Visual Designs through Transformer-Based Architectures and Specialized Datasets" will be presented by Tommaso on June 26 during Paper Session 3 - Modelling and Development at 15:30.

This paper addresses the limitations of current methods for translating web designs into code, which often rely on simplistic datasets and have not fully explored the potential of transformer-based models. Our contributions include:

  1. A web scraping methodology to collect and process a diverse dataset of real-world websites.
  2. A synthetic dataset of webpage mockups and their sketched conversions.
  3. An evaluation of two recent multimodal transformer architectures on these datasets.

Results on synthetic and sketch-based datasets demonstrate the architectures' potential as effective design-to-code automation solutions, while also identifying remaining challenges in modeling real-world website complexity.

Bridging Web and Figma: Automating Large-Scale UI Dataset Generation for AI-Enhanced Design​

The second paper, "Bridging Web and Figma: Automating Large-Scale UI Dataset Generation for AI-Enhanced Design" will also be presented by Francesca on June 27 during Paper Session 9 - Design Support at 13:30 as a Late-Breaking Result.

In this work, we introduce an automated pipeline capable of converting any HTML-based user interface into a structured, Figma-compatible JSON representation. This pipeline was applied to WebUI, a large-scale HTML-based dataset, to capture structural and semantic details.

To validate its effectiveness, five state-of-the-art layout generation models were trained on data generated by our pipeline and compared with models trained on the manually annotated Rico dataset. Experimental results demonstrate that these models achieve comparable performance, suggesting our pipeline can produce high-quality data suitable for training AI models that integrate into design workflows.

All conference proceedings, including late-breaking results, are published by ACM, with full papers appearing in the PACM EICS journal series.

Additional information: