Research map / Vision-language and multimodal / Human and embodied modeling
Visual Pinwheel Centers Act as Geometric Saliency Detectors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds a 2D self-evolving spiking neural network with Hebbian-like plasticity and anatomical overlap data, showing that pinwheel-center neurons respond faster and more selectively to geometric complexity than salt-and-pepper neurons.
Problem
The functional advantage of pinwheel organization over salt-and-pepper organization in V1 orientation maps is unclear, and it is unknown how pinwheel centers process complex contours and saliency compared with iso-orientation domains.
Approach
A 2D E/I leaky integrate-and-fire network trained on whitened natural images with Hebbian-Oja and correlation-measuring rules; varying the receptive-field input overlap produces salt-and-pepper or pinwheel maps, and the pinwheel-center neurons are then shown to respond faster and more saliently to contour complexity.
Key results
- SESNN overlap percentage correlates with anatomical overlap across mouse, cat, and macaque (R2 = 0.97), and the model yields pinwheel density 3.175 ± 0.397 vs ~3.327 in macaque.
- Saliency vs. complexity correlation on BSDS500 is R2 = 0.82 (y = 1.39x − 0.55).
- Pinwheel-center neurons respond with significantly shorter onset latency to complex contours than salt-and-pepper neurons (p<0.0001, Welch's t-test).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Increase visual input overlap | SESNN model · V1 orientation maps | pinwheel count | Hurt |
| Increase visual input overlap | SESNN model · V1 orientation maps | NNPD | Helped |
| Increase visual input overlap | SESNN model · V1 orientation maps | hypercolumn size | Helped |
| Use SESNN model instead of E-I baseline | Macaque pinwheel modeling · Macaque pinwheels | pinwheel density (pinwheels/Λ²): ∼2.941 -> 3.175 ± 0.397 | Mixed |
| Use pinwheel organization instead of salt-and-pepper organization | SESNN model · BSDS500 and star-like contour stimuli | response onset latency | Helped |
| Use pinwheel organization instead of salt-and-pepper organization | SESNN model · BSDS500 and star-like contour stimuli | response to contour complexity | Helped |
| Increase local pixel entropy | BSDS500 images · BSDS500 | saliency: R2 = 0.82; y = 1.39*x - 0.55 | Helped |
| Increase local pixel entropy | Star-like shapes · Star-like shape stimuli | saliency: R2 = 0.80; y = 1.37*x - 0.52 | Helped |
| Disrupt local connectivity (ablation) | SESNN pinwheel orientation map · V1 pinwheel orientation map | complexity | Hurt |
| Shuffle the spatial arrangement of orientation-tuned RFs (ablation) | SESNN pinwheel orientation map · V1 pinwheel orientation map | complexity | Hurt |
| Shuffle feed-forward and lateral connections (ablation) | SESNN pinwheel orientation map · V1 pinwheel orientation map | complexity | Hurt |
| Increase neuronal connection range | SESNN model · V1 orientation maps | pinwheel count | Hurt |
Limitations
- Geometric-complexity measure relies on local pixel entropy, which the authors acknowledge cannot directly capture geometric shapes; a polygon-based geometric entropy is only explored in the appendix.
- Analysis is purely simulation-based with no direct biological experiment validating the predicted saliency role of pinwheel centers.
- Code and data available only on request, limiting reproducibility.
Experimental setup
Datasets: BSDS500 anatomical V1/retina data from cats, macaques, mice (Tables 2-3)
Benchmarks: BSDS500 boundary/saliency maps macaque pinwheel statistics
Metrics: pinwheel density nearest-neighbor pinwheel distance hypercolumn size orientation bimodal ratio response onset latency local pixel entropy correlation with saliency
Baselines: SNN-based model (Bauer et al. baseline) E-I baseline model
Builds on: Hebbian-Oja rule correlation measuring rule leaky integrate-and-fire neurons
Closest papers by meaning
- Emergent Orientation Maps —— Mechanisms, Coding Efficiency and Robustness (ICLR 2025 · similarity 0.93)
- Connectome-Based Modelling Reveals Orientation Maps in the Drosophila Optic Lobe (NeurIPS 2025 · similarity 0.86)
- Finding Visual Saliency in Continuous Spike Stream (AAAI 2024 · similarity 0.84)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.84)
- NEvo: Neural-Guided Evolutionary Video Synthesis for Dynamic Visual Selectivity (NeurIPS 2026 · similarity 0.84)
- A Computational Framework for Modeling Emergence of Color Vision in the Human Brain (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.