Hidden Layers: An Interactive Installation for Exploring the Neural Semantics of Image Synthesis
Figure 1: A visualization of the hidden layers of our GAN-based prototype generating fashion imagery.Normally concealed within the black box of a generative model, this visualization reveals the activation of the hidden layers-progressing from left to right-consisting of many channels, here visualized as grids of neurons.Magenta pixels indicate positive, while cyan pixels represent negative activation of a neuron.We explore their material properties as a new modality for interaction.
Authors
- David Bau (ORCID: https://orcid.org/0000-0003-1744-6765)
- Jaden Fiotto-Kaufman
- Imke Grabe (ORCID: https://orcid.org/0000-0002-5122-6769)
- Tom L. Jenkins (ORCID: https://orcid.org/0000-0001-5774-0278)
Institutions
- Northeastern University (US)
- IT University of Copenhagen (DK)
Publication Details
- Published
- 2025-08-18
- DOI
- https://doi.org/10.1145/3737609.3747108
- Primary Topic
- Image Retrieval and Classification Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00