Three recent talks from IVRL
© 2026 EPFL
Prof. Sabine Süsstrunk recently delivered three talks featuring the work of IVRL, spanning neural cellular automata (NCAs), generative models, and color and noise in computational photography.
"Neural Cellular Automata", Swiss Vision Symposium, EPFL, 1 September 2026
Prof. Süsstrunk presented IVRL's work on Neural Cellular Automata (NCAs) at the Swiss Vision Symposium, held at EPFL. The event brought together Swiss researchers in retinal biology, clinical vision science, visual neuroscience, and computational neuroscience. The talk was part of the computational neuroscience session. NCAs are models in which each tiny agent runs the same small neural network and updates its state from its local neighborhood only, so that complex global patterns emerge and self-organize from simple local rules.
IVRL's work on NCAs includes:
- DyNCA: Real-time Dynamic Texture Synthesis Using Neural Cellular Automata (CVPR 2023),
- Emergent Dynamics in Neural Cellular Automata (ALIFE 2024),
- NoiseNCA: Noisy Seed Improves Spatio-Temporal Continuity of Neural Cellular Automata (ALIFE 2024),
- Mesh Neural Cellular Automata (SIGGRAPH 2024),
- Discovering Partial Differential Equations with Neural Cellular Automata (Artificial Life, 2026),
- Neural Particle Automata: Learning Self-Organizing Particle Dynamics (SIGGRAPH 2026), and
- Neural Cellular Automata: From Cells to Pixels (SIGGRAPH 2026).
"On Generating Image and Video Hallucinations", Paul Scherrer Institute (PSI), 18 September 2026
Prof. Süsstrunk presented an updated version of her talk "On Generating Image and Video Hallucinations". AI model hallucinations can be harmful when presenting wrong information as fact, but they can also be useful whenever the goal is a visually plausible image rather than a physically accurate one, as in photography and content creation.
The talk gave an overview of the main paradigms of generative AI (GANs, autoregressive models, synthesis through optimization, and diffusion and flow-matching models) and presented various IVRL works on the topic:
- Diffusion in Style (ICCV 2023),
- Exploiting the Signal-Leak Bias in Diffusion Models (WACV 2024),
- VibrantLeaves: A principled parametric image generator for training deep restoration models (to appear in TIP 2026),
- Kernel Modeling Super-Resolution on Real Low-Resolution Images (ICCV 2019),
- Dark Noise Diffusion: Noise Synthesis for Low-Light Image Denoising (ICCP 2025, TPAMI 2025),
- 2-Shots in the Dark: Low-Light Denoising with Minimal Data Acquisition (CVPR 2026),
- BIGPrior: Towards Decoupling Learned Prior Hallucination and Data Fidelity in Image Restoration (TIP 2022),
- Covariance Mismatch in Diffusion Models (preprint, 2024), and
- Frequency-Aware Denoising Score for Text-Guided Latent Diffusion Image Editing (CVPR 2025).
"Color & Noise", Computational Photography Conference at the 2026 HUAWEI Future Device Technology Alps Summit in Vienna, Austria, 23-25 September 2026
Prof. Süsstrunk presented three recent IVRL works on the topic of color and noise at the 2026 HUAWEI Future Device Technology Alps Summit in Vienna, Austria.
The talk covered IVRL's work on Color Constancy in Hyperspectral Imaging via Reduced Spectral Spaces (ICIP 2026), Dark Noise Diffusion: Noise Synthesis for Low-Light Image Denoising (ICCP 2025, TPAMI 2025), and 2-Shots in the Dark: Low-Light Denoising with Minimal Data Acquisition (CVPR 2026).