Silvia Bracco, Matteo Sipione
DreamWalking: Decoding Imagined Objects from EEG into Interactive 3D through a Co-Imagining Loop.
Rel. Andrea Sanna, Federico Manuri. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Del Cinema E Dei Mezzi Di Comunicazione, 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (20MB) | Preview |
Abstract
Reconstructing visual experience from brain activity has advanced rapidly, but most work relies on fMRI and produces flat, single-shot 2D images. This thesis presents DreamWalking, to the best authors' knowledge the first system to turn non-invasive EEG into interactive 3D objects, and to treat decoding not as one-way reconstruction but as a co-imagining loop between user and machine. EEG is portable and inexpensive but noisy and low in spatial resolution, and electrode montages differ widely across devices. This work introduces an EEG encoder designed to generalise across these montages, trained to produce a DINOv2-like embedding from brain activity. Rather than feeding this vector directly into a generator, the system uses it as a query: a retrieval stage computes cosine similarity against a FAISS index over millions of ImageNet-21k image embeddings, anchoring the noisy prediction to a real concept.
The retrieved image then drives an image-to-3D generator (Hunyuan3D 2.1) to produce a mesh
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
