• Installation
• Usage
• Render event data
• Render frames

Installation

1. Clone this repository (git clone https://github.com/neuromorphicsystems/davis346-recorder) or download and unzip a copy (<> Code button). Open a terminal and cd to the repository's directory.

2. Create a virtual environment

   # Linux, macOS
   python3 -m venv .venv
   
   # Windows
   python -m venv .venv
   

3. Activate the virtual environment

   # Linux, macOS
   source .venv/bin/activate
   
   # Windows (cmd.exe)
   .venv\Scripts\activate.bat
   
   # Windows (PowerShell)
   .venv\Scripts\Activate.ps1
   

4. Install dependencies

   pip install neuromorphic-drivers faery PySide6 pillow
   

Usage

1. Activate the virtual environment (skip this if you just went through the installation steps)

   # Linux, macOS
   source .venv/bin/activate
   
   # Windows (cmd.exe)
   .venv\Scripts\activate.bat
   
   # Windows (PowerShell)
   .venv\Scripts\Activate.ps1
   

2. Run the application

   python davis346_recorder.py
   

See https://docs.inivation.com/hardware/hardware-advanced-usage/biasing.html for a guide on bias tuning.

Render event data

1. Create a directory called recordings

2. Move event files of interest (.csv) in the directory recordings

3. Run faery init

4. Faery cannot automatically tell the sensor width and height from a CSV file. In faery_script.py (created by the previous command), find all the calls to faery.events_stream_from_file and add the argument dimensions_fallback=(346, 260). The modified calls should look as follows:

   faery.events_stream_from_file(input, dimensions_fallback=(346, 260))
   

5. Run faery run

See https://aestream.github.io/faery/batch/ for documentation and an example on how to generate slow-motion videos.

Render frames

The Davis346 ADC has a precision of 10 bits, hence the frame values are in the range [0, 1023].

Neuromorphic-drivers multiplies the raw values by 64, making the effective range [0, 65472], to avoid display issues (16 bits images look very dark if only the range [0, 1023] is used).

The frames' timestamps (same clock as the events) are stored in frames_metadata.csv. The raw frames (native endian u16) are stored in frames/dddddd.raw. They can be loaded in Python as follows.

# read the frame as a numpy array
import numpy

with open("data/<recording name>/frames/000000.raw", "rb") as frame_file:
    array = numpy.frombuffer(frame_file.read(), dtype=numpy.uint16).reshape((260, 346))

# save the frame a 16-bit PNG image
import PIL.Image

PIL.Image.fromarray(array).save("test.png")