CARLA – Computer Vision Approaches in Low Annotation Regimes: biomedical and facial image analysis (2022-2025)

The CARLA project — “Métodos de visión artificial en entornos de pocas anotaciones: análisis de imagen biomédica y facial” — focuses on the development of AI methods for visual recognition in low-data environments, where annotated datasets are scarce or expensive to obtain.

Led by the Computer Vision and Pattern Discovery (CVPD) group at the University of the Basque Country (UPV/EHU), CARLA targets two key application domains:

• Biomedical image analysis: Enhancing diagnostic tools and research through robust AI-based segmentation and classification under minimal annotation constraints.
• Facial image analysis: Improving recognition and aesthetics assessment tasks with annotation-efficient deep learning techniques.

🔬 Scientific Goals

The CARLA project focuses on the development of advanced computer vision and machine learning techniques to address two core challenges:

• The need for large amounts of annotated data for training state-of-the-art models.
• The limited generalization capacity of these models when applied to new but related domains.

Its specific goals are:

• Develop deep learning techniques capable of operating in low-annotation settings by leveraging data characteristics, feature representations, and task-specific regularization strategies.

• Explore domain adaptation and domain generalization approaches to improve the transferability of models across different but related visual domains (e.g., from one biomedical imaging modality to another).

• Integrate and validate these techniques on two application areas:

  • Biomedical image analysis, focusing on segmentation and classification tasks.
  • Facial image analysis, particularly on face interpretation and aesthetic evaluation.

• Deliver tools and pipelines that can be integrated into open-source frameworks, promoting reproducibility, usability, and transfer of methods to real-world scenarios.

The project combines theoretical research with practical experimentation using both synthetic datasets and real annotated collections in the biomedical and facial imaging domains.

👥 Project Team

• Principal Investigators:

  • Dr. Fadi Dornaika (Ikerbasque Research Professor, UPV/EHU)
  • Dr. Ignacio Arganda-Carreras (Ikerbasque Research Associate Professor, UPV/EHU, DIPC, Biofisika Institute)

• Research Team:

  • Dr. Nagore Barrena (Assistant Professor, UPV/EHU)
  • Dr. Urtzi Ayesta (Ikerbasque Research Professor, UPV/EHU)

• Work Team:

  • Dr. Fares Bougourzi (Researcher, University of Salento, Italy)
  • Dr. Alireza Bosaghzadeh (Associate Professor, Shahid Rajaee Teacher Training University, Iran)
  • Dr. Donglai Wei (Associate Professor, Boston College, USA)
  • Daniel Franco-Barranco (PhD student, Donostia International Physics Center)

🕒 Timeline

• Start: September, 1 2022
• End: August, 31 2025

💰 Funding

• Funded by the Spanish Ministry of Science and Innovation (MCIN) under the Generación de Conocimiento 2021 (Proyectos de Investigación Orientada) program.
• Total funding: 101,640€ (including indirect costs)
• Reference code: PID2022-137580OB-I00