FaCells: AI's Artistic Eye - Visualizing Face Attributes Through Interpretable Line Sketches
By Xavier Ignacio Gonzalez
Published on November 24, 2025| Vol. 1, Issue No. 1
Content Source
This is a curated briefing. The original article was published on cs.CV updates on arXiv.org.
Summary\
FaCells is an innovative method and art exhibition that transforms aligned facial photographs from the CelebA dataset into vector sketches, designed for XY plotters. The project trains a bidirectional LSTM to predict 40 facial attributes, introducing a novel architectural modification to derive "per-point attribute scores." By aggregating points that exceed specific attribute thresholds, FaCells creates statistical abstractions (e.g., for "Eyeglasses" or "Wavy Hair") that serve as interpretable, line-based artworks. The project emphasizes interpretability as a creative tool, generating reproducible, physically present art while also acknowledging dataset biases and the limitations of labels, effectively bridging data, AI models, and physical art.
\
Why It Matters\
FaCells presents a compelling convergence of AI interpretability, computer vision, and artistic expression, offering profound implications for the AI industry. Firstly, it champions a novel approach to model understanding, moving beyond abstract metrics or feature maps to produce tangible, visual explanations. By translating complex internal model states into line-based art, FaCells makes AI's "thinking" process more accessible and intuitive, allowing professionals to literally "see" which parts of a sketch contribute to a specific attribute prediction. This innovative visualization technique can foster greater trust and transparency in AI systems, especially in sensitive areas like facial analysis where understanding biases is paramount. Secondly, it elevates AI art beyond mere aesthetic generation, positioning it as a powerful tool for scientific inquiry and ethical discussion. The project's explicit acknowledgment of dataset biases, integrated into the artistic output, encourages a deeper, more critical engagement with AI's inherent limitations and societal impacts. This multidisciplinary fusion not only pushes the boundaries of AI interpretability research but also demonstrates how artistic creativity can serve as a potent medium for demystifying AI, fostering public understanding, and guiding the development of more responsible and human-centric AI technologies.