talon-io 0.0.1

TALON IO: Streaming I/O, event processing, and sensor integration for TALON hardware

TALON I/O

Streaming I/O, event processing, and neural encoding for TALON neuromorphic systems. Provides vectorized event streaming, spike encoding, and format conversion for spike-based inference pipelines.

Installation

uv add talon-io

Optional extras for vendor camera file import:

uv add talon-io[metavision]  # EVT2/EVT3 file import
uv add talon-io[tonic]       # Numba-accelerated transforms

Features

• Event Streaming: Buffered zero-copy event reading (23M+ events/sec)
• Neural Encoding: Rate (Poisson), temporal (time-to-first-spike), delta, latency encoding
• Frame Conversion: Events to dense frames and spike rasters (vectorized)
• Multi-Sensor Sync: Temporal alignment across multiple event sources
• Network Streaming: UDP event receiver for real-time hardware inference
• Vendor Format Import (optional): EVT2/EVT3, AEDAT4 file adapters

Quick Start

from talon.io import streaming, encoding, formats
import numpy as np

# Stream events from binary file
reader = streaming.BufferedEventReader("events.bin", chunk_size=10000)
for chunk in reader:
    print(f"Chunk: {len(chunk)} events")

# Encode static images as spike trains for SNN inference
image = np.random.rand(28, 28).astype(np.float32)
spikes = encoding.rate_encode(image, n_steps=100, rate_scale=1.0)
print(f"Spike train shape: {spikes.shape}")

# Convert events to dense frames
events = streaming.generate_random_events(50000, sensor_size=(640, 480))
frames = formats.events_to_frames(events, sensor_size=(640, 480), n_frames=10)
print(f"Frames shape: {frames.shape}")

License

MIT