Datasets:

Agnuxo
/

optical-neuromorphic-eikonal-benchmarks

case stringclasses 5 values	gpu_time float64 0 0	cpu_time float64 0.07 1.16	speedup float64 32.2 276	mae float64 0.01 0.01	path_ratio float64 1 1.04	metadata stringclasses 5 values
complex_512.npz	0.004199	1.158727	275.98	0.006943	1.041743	{'dataset': 'Synthetic', 'name': 'complex_512', 'size': 512, 'obstacle_density': 0.3, 'speed_mode': 'random', 'seed': 7}
gradient_256.npz	0.002881	0.283779	98.5	0.006797	1.042938	{'dataset': 'Synthetic', 'name': 'gradient_256', 'size': 256, 'obstacle_density': 0.2, 'speed_mode': 'gradient', 'seed': 5}
maze_511.npz	0.004011	0.738994	184.24	0.007921	1.021085	{'dataset': 'Synthetic', 'name': 'maze_511', 'size': 511, 'obstacle_density': 0.3, 'speed_mode': 'maze', 'seed': 11}
medium_256.npz	0.003247	0.272064	83.8	0.005396	1.015287	{'dataset': 'Synthetic', 'name': 'medium_256', 'size': 256, 'obstacle_density': 0.2, 'speed_mode': 'uniform', 'seed': 3}
sparse_128.npz	0.002294	0.073909	32.21	0.005169	1.004002	{'dataset': 'Synthetic', 'name': 'sparse_128', 'size': 128, 'obstacle_density': 0.1, 'speed_mode': 'uniform', 'seed': 1}

Optical Neuromorphic Eikonal Solver - Benchmark Datasets

Overview

Benchmark datasets for evaluating the Optical Neuromorphic Eikonal Solver, a GPU-accelerated pathfinding algorithm achieving 30-300× speedup over CPU Dijkstra.

🎯 Key Results

134.9× average speedup vs CPU Dijkstra
0.64% mean error (sub-1% accuracy)
1.025× path length (near-optimal paths)
2-4ms per query on 512×512 grids

📊 Dataset Content

5 synthetic pathfinding test cases covering diverse scenarios:

File	Grid Size	Cells	Obstacles	Speed Field	Difficulty
sparse_128.npz	128×128	16,384	10%	Uniform	Easy
medium_256.npz	256×256	65,536	20%	Uniform	Medium
gradient_256.npz	256×256	65,536	20%	Gradient	Medium
maze_511.npz	511×511	261,121	30% (maze)	Uniform	Hard
complex_512.npz	512×512	262,144	30%	Random	Hard

Plus: benchmark_results.csv with performance metrics

📋 Format

Each .npz file contains:

{
    'obstacles': np.ndarray,  # (H,W) float32, 1.0=blocked, 0.0=free
    'speeds': np.ndarray,     # (H,W) float32, propagation speed
    'source': np.ndarray,     # (2,) int32, [x,y] start coordinates
    'target': np.ndarray,     # (2,) int32, [x,y] goal coordinates
    'metadata': str           # JSON with provenance info
}

🔧 Loading Data

import numpy as np
from huggingface_hub import hf_hub_download

# Download dataset
file_path = hf_hub_download(
    repo_id="Agnuxo/optical-neuromorphic-eikonal-benchmarks",
    filename="maze_511.npz",
    repo_type="dataset"
)

# Load
data = np.load(file_path, allow_pickle=True)
obstacles = data['obstacles']
speeds = data['speeds']
source = tuple(data['source'])
target = tuple(data['target'])

print(f"Grid: {obstacles.shape}")
print(f"Start: {source}, Goal: {target}")

🎮 Interactive Demo

Try the interactive pathfinding demo: Space Link

📄 Paper & Code

Paper: GitHub
Code: GitHub Repository
Author: Francisco Angulo de Lafuente

📚 Citation

@misc{angulo2025optical,
  title={Optical Neuromorphic Eikonal Solver Benchmark Datasets},
  author={Angulo de Lafuente, Francisco},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/Agnuxo/optical-neuromorphic-eikonal-benchmarks}
}