TINTOlib is a Python library that transforms tabular data into synthetic images, enabling the use of Deep Learning models such as CNNs and Vision Transformers on tabular data.

Which deep learning models can I use with TINTOlib?

TINTOlib supports architectures like CNNs, Vision Transformers (ViTs), and hybrid neural networks that combine synthetic images with classic tabular features.

Why convert tabular data into images?

Converting tabular data into images allows leveraging computer vision models for tabular data tasks, often improving accuracy and generalization.

Where can I learn to use TINTOlib?

You can access the free course on Udemy, follow the bilingual video tutorial on GitHub, and consult the full documentation on ReadTheDocs. Links are provided on this page.

TINTOlib — From Tabular Data to Synthetic Images for Deep Learning

TINTOlib — Tabular Data → Synthetic Images

Open-source Python library that transforms tabular data into synthetic images, enabling the use of vision-based models such as CNNs, ViTs, and hybrid architectures.

Documentation GitHub

🎉 Learn TINTOlib for Free

📚 Udemy

Free course with videos and practical examples.

Access Udemy

📺 GitHub

Bilingual video tutorials + notebooks.

🎬 TINTOlib — Overview Video

🌍 Talk Map: TINTOlib Around the World

Key conferences and seminars where TINTOlib has been presented or applied.

🧠 Research and Software Publications

📄 Research Articles

Manuel Castillo-Cara et al. MIMO-Based Indoor Localisation with Hybrid Neural Networks. IEEE JSTSP. DOI: 10.1109/JSTSP.2025.3555067
Reewos Talla-Chumpitaz, Manuel Castillo-Cara et al. Blurring Image Techniques for Bluetooth-based Indoor Localisation. Information Fusion. DOI: 10.1016/j.inffus.2022.10.011

💾 Software Articles

Manuel Castillo-Cara et al. TINTO: Converting Tidy Data into Images. SoftwareX. DOI: 10.1016/j.softx.2023.101391

📊 Benchmark of TINTOlib vs Classical Models

This benchmark compares TINTOlib transformation methods and neural architectures against traditional and ensemble models on multiple datasets. You can explore the interactive version directly on the official site.

🌐 Visit TINTOlib Benchmark Page

🧪 Supported Methods in TINTOlib

Method	Description
TINTO	Converts tidy tabular data into synthetic images using several layout algorithms and optional image preprocessing.
IGTD	Places features on a grid based on correlations, producing synthetic images that preserve variable relationships.
REFINED	Optimizes the 2D placement of variables to enhance feature locality for CNN processing.
BarGraph	Represents each sample as a bar-chart image of its feature values for direct CNN classification.
DistanceMatrix	Builds an image from pairwise feature distances, revealing similarity patterns in a matrix form.
Combination	Fuses multiple image encodings into a single input, merging complementary representations.
SuperTML	Transforms tabular data into text-based images, embedding values directly as text.
FeatureWrap	Encodes features in circular/spiral patterns to emphasize variable ordering and grouping.
BIE	Generates compact, binary-encoded images from tabular features for classification.

🎓 License

TINTOlib is released under the Apache License 2.0.

Ph.D. Manuel Castillo-Cara