Invited Talk

"Restoring What the Flood Took: AI for Personal Memory Reconstruction"
by Cèsar Ferri

In October 2024, a cut-off low (DANA) caused severe flooding in the Valencian Community, in eastern Spain, leading to the widespread loss and deterioration of personal and documentary photographs. In response, several institutions and universities launched Recuperar les memòries, a collaborative initiative aimed at preserving both personal and collective memory in the face of climate-related disasters. The project combines manual photo restoration with artificial intelligence techniques to recover damaged images and help individuals reconnect with their visual histories.

This talk presents the AI-based system developed as part of the initiative to detect, segment, and reconstruct flood-damaged photographs. The architecture follows a two-stage pipeline: damage identification and generative inpainting. Damage is first localized through image segmentation, with post-processing techniques applied to refine mask quality. These regions are then reconstructed using inpainting methods, with tuning of generation parameters and the use of negative prompts to reduce unrealistic artifacts. A dedicated face reconstruction module was integrated to improve the fidelity of portraits and preserve identity.

The final result is a functional, accessible restoration tool with a simple interface, capable of handling both individual and batch processing. This work demonstrates the practical and emotional value of AI in post-disaster recovery, offering a way to restore not only images, but memory itself.