Welcome to the official repository of the Tech4Diet Research Lab at the University of Alicante. We focus on applied research in Artificial Intelligence, particularly in areas such as machine learning and computer vision. This repository hosts resources, code, datasets, and documentation from our ongoing projects.
- 26/03/2025 – 🗞️ First public release of the GLORiA Project!
The initial version of GLORiA is now available on GitHub. This release includes deep learning models for fish classification, visual explainability tools and a new dataset for detecting escaped fish from aquaculture facilities. Stay tuned for upcoming modules and releases! - 27/06/2025 – 🎤 M1+ presented at SARTECO 2025!
The classification system based on text–image embeddings and deep learning CNN models was presented during the national conference on advanced technologies in computing.
🔗 View the LinkedIn post: Showcasing GLORiA at SARTECO 2025
- 📰 GLORIA Project – Official Pleamar Page: GLORIA Tools – Aquaculture tools for long-term sustainability
- 📰 University of ALicante: Exposición del Programa Pleamar en la UA
- 📰 Programa Pleamar – News: Tres proyectos de investigación ambiental en acuicultura
📘 Benchmarking Deep Learning Models for Fish Classification
- Image segmentation and enhancement of the dataset
- Loss function design and augmentation strategies for class imbalance
- Fine-tuning of baseline CNN models
- Fine-tuning of baseline Vision Transformer (ViT) models
- CLIP-based zero-shot and prompt-driven classification
- Comparative analysis of model performance across approaches
- 🔗 Related repository: [TFG 3 Class Classification]
📗 Explainability and Model Transparency
- Prompt refinement to enhance model interpretability
- Extraction of key visual features used by the models
- Integration of interpretability techniques (e.g., Grad-ECLIP, t-SNE, manual feature manipulation)
- Comparison between model-derived features and expert annotations
- Application of explainability pipeline to escaped fish detection scenarios
- 🔗 Related repository: [GLORiA-M1+ Explainability (Pending update)]
📕 Dataset Expansion and Open Science
- Inclusion of new high-quality laboratory images
- Expansion of the dataset to include more complex, non-optimal conditions
- Annotation and curation of edge cases and challenging specimens
- Release of a public version of the extended dataset with full documentation
- 🔗 Related repository: [GLORiA-Dataset (Pending update)]
-
Jerez, M. et al. (2024). GLORiA: Automatic Identification of Fish Species and Their Farmed or Wild Origin by Computer Vision and Deep Learning
📚 Springer Link -
Jerez, M. et al. (2025). Comparative Study of Deep Learning Approaches for Fish Origin Classification
📚 Springer Link -
Jerez, M. et al. (2026). Domain-Aware Foundation Vision-Language Models for Explainable Identification of Wild and Farmed Fish
📚 Springer Link Pending -
Jerez, M. et al. (2026). The GLORiA fish farm escapes identification dataset
📚 Springer Link Pending
| Name | Role | GitHub | Contact |
|---|---|---|---|
| Dr. Andrés Fuster Guilló | Principal Investigator | – | [email protected] |
| Dr. Jorge Azorín López | Principal Investigator | – | [email protected] |
| Dr. Marcelo Saval Calvo | Principal Investigator | – | [email protected] |
| Dr. Nahuel Emiliano Garcia d'Urso | Principal Investigator | @nawue | [email protected] |
| Bernabé Sanchez Sos | PhD Student | @Bernabe19 | [email protected] |
| Ismael Beviá Ballesteros | PhD Student | @ibevias | [email protected] |
| Mario Jerez Tallón | Research Assistant | @Mariojt72 | [email protected] |
