TINTOlib: A Python library for transforming tabular data into synthetic images for deep neural networks
Transforming tabular data into synthetic images enables the application of vision-based deep learning models – such as Convolutional Neural Networks and Vision Transformers – to non-visual tasks. This paper presents TINTOlib, the first Python library to unify a diverse set of tabular data into synth...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad Nacional de Educación a Distancia |
| Repositorio: | e-spacio. Repositorio Institucional de la UNED |
| Idioma: | inglés |
| OAI Identifier: | oai:e-spacio.uned.es:20.500.14468/31014 |
| Acceso en línea: | https://hdl.handle.net/20.500.14468/31014 |
| Access Level: | acceso abierto |
| Palabra clave: | 1203.04 Inteligencia artificial Hybrid neural networks Synthetic images TINTOlib Tabular-to-image Tabular2image |
| Sumario: | Transforming tabular data into synthetic images enables the application of vision-based deep learning models – such as Convolutional Neural Networks and Vision Transformers – to non-visual tasks. This paper presents TINTOlib, the first Python library to unify a diverse set of tabular data into synthetic image transformation methods into a cohesive, extensible framework. TINTOlib unifies parametric and non-parametric tabular to synthetic image methods within a consistent interface, lowering the barrier to apply, compare, and extend these techniques. The generated images can be directly used with vision models or integrated into Hybrid Neural Networks that combine visual and tabular branches. By addressing reproducibility, scalability, and modularity, the library simplifies experimentation and deployment of deep learning pipelines on tabular data. Illustrative results show that the use of synthetic images can achieve competitive or superior performance compared to state-of-the-art classical models in both regression and classification tasks, with outcomes varying across transformation techniques and architectural backbones. This underscores the utility of TINTOlib in bridging tabular data with vision-based deep learning via synthetic image representations. |
|---|