.. _paper-cgil:

CLIP with Generative Latent Replay: a Strong Baseline for Incremental Learning
==============================================================================

    - :fa:`circle-check` `Proceedings of the 35th British Machine Vision Conference <https://bmvc2024.org/proceedings/863/>`_
    - :fa:`calendar` November 2024
    - :fa:`scroll` `BMVC <https://bmvc2024.org/>`_
    - :fa:`tags` :bdg-primary:`Continual Learning` :bdg-primary:`Multi-Modal` :bdg-primary:`Generative Replay` :bdg-primary:`Zero-Shot Continual`

This work introduces **CGIL**, a simple baseline designed to address the challenge of learning from a stream of data while preserving the *zero-shot* capabilities of Vision-Language models. Our approach leverages feature-level *generative replay* to learn class-specific prompts, enabling the adaptation of the textual encoder without compromising zero-shot performance.