AEDAT is a fast AEDAT 4 python reader, with a Rust underlying implementation.

Run pip install aedat to install it.

• Example
• Process frames
  • Pillow (PIL)
  • OpenCV
• Detailed example
• Install from source
  • Linux
  • macOS
  • Windows
• Contribute
• Publish

Example

The aedat library provides a single class: Decoder. A decoder object is created by passing a file name to Decoder. The file name must be a path-like object.

import aedat

decoder = aedat.Decoder("/path/to/file.aedat")
print(decoder.id_to_stream())

for packet in decoder:
    print(packet["stream_id"], end=": ")
    if "events" in packet:
        print("{} polarity events".format(len(packet["events"])))
    elif "frame" in packet:
        print("{} x {} frame".format(packet["frame"]["width"], packet["frame"]["height"]))
    elif "imus" in packet:
        print("{} IMU samples".format(len(packet["imus"])))
    elif "triggers" in packet:
        print("{} trigger events".format(len(packet["triggers"])))

Process frames

Pillow (PIL)

import aedat
import PIL.Image # https://pypi.org/project/Pillow/

index = 0
for packet in decoder:
    if "frame" in packet:
        image = PIL.Image.fromarray(
            packet["frame"]["pixels"],
            mode=packet["frame"]["format"],
        )
        image.save(f"{index}.png")
        index += 1

OpenCV

import aedat
import cv2 # https://pypi.org/project/opencv-python/

index = 0
for packet in decoder:
    if "frame" in packet:
        image = packet["frame"]["pixels"]
        if packet["frame"]["format"] == "RGB":
            image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
        elif packet["frame"]["format"] == "RGBA":
            image = cv2.cvtColor(image, cv2.COLOR_RGBA2BGRA)
        cv2.imwrite(f"{index}.png", image)
        index += 1

Detailed example

This is the same as the first example, with detailed comments:

import aedat

decoder = aedat.Decoder("/path/to/file.aedat")
"""
decoder is a packet iterator with an additional method id_to_stream
id_to_stream returns a dictionary with the following structure:
{
    <int>: {
        "type": <str>,
    }
}
type is one of "events", "frame", "imus", "triggers"
if type is "events" or "frame", its parent dictionary has the following structure:
{
    "type": <str>,
    "width": <int>,
    "height": <int>,
}
"""
print(decoder.id_to_stream())

for packet in decoder:
    """
    packet is a dictionary with the following structure:
    {
        "stream_id": <int>,
    }
    packet also has exactly one of the following fields:
        "events", "frame", "imus", "triggers"
    """
    print(packet["stream_id"], end=": ")
    if "events" in packet:
        """
        packet["events"] is a structured numpy array with the following dtype:
            [
                ("t", "<u8"),
                ("x", "<u2"),
                ("y", "<u2"),
                ("on", "?"),
            ]
        """
        print("{} polarity events".format(len(packet["events"])))
    elif "frame" in packet:
        """
        packet["frame"] is a dictionary with the following structure:
            {
                "t": <int>,
                "begin_t": <int>,
                "end_t": <int>,
                "exposure_begin_t": <int>,
                "exposure_end_t": <int>,
                "format": <str>,
                "width": <int>,
                "height": <int>,
                "offset_x": <int>,
                "offset_y": <int>,
                "pixels": <numpy.array(shape=(height, width), dtype=uint8)>,
            }
        format is one of "L", "RGB", "RGBA"
        """
        print("{} x {} frame".format(packet["frame"]["width"], packet["frame"]["height"]))
    elif "imus" in packet:
        """
        packet["imus"] is a structured numpy array with the following dtype:
            [
                ("t", "<u8"),
                ("temperature", "<f4"),
                ("accelerometer_x", "<f4"),
                ("accelerometer_y", "<f4"),
                ("accelerometer_z", "<f4"),
                ("gyroscope_x", "<f4"),
                ("gyroscope_y", "<f4"),
                ("gyroscope_z", "<f4"),
                ("magnetometer_x", "<f4"),
                ("magnetometer_y", "<f4"),
                ("magnetometer_z", "<f4"),
            ]
        """
        print("{} IMU samples".format(len(packet["imus"])))
    elif "triggers" in packet:
        """
        packet["triggers"] is a structured numpy array with the following dtype:
            [
                ("t", "<u8"),
                ("source", "u1"),
            ]
        the source value has the following meaning:
            0: timestamp reset
            1: external signal rising edge
            2: external signal falling edge
            3: external signal pulse
            4: external generator rising edge
            5: external generator falling edge
            6: frame begin
            7: frame end
            8: exposure begin
            9: exposure end
        """
        print("{} trigger events".format(len(packet["triggers"])))

Because the lifetime of the file handle is managed by Rust, decoder objects are not compatible with the with statement. To ensure garbage collection, point the decoder variable to something else, for example None, when you are done using it:

import aedat

decoder = aedat.Decoder("/path/to/file.aedat")
# do something with decoder
decoder = None

Install from source

This library requires Python 3.x, x >= 5, and NumPy. This guide assumes that they are installed on your machine.

Note for Windows users: this library requires the x86-64 version of Python. You can download it here: https://www.python.org/downloads/windows/ (the default installer contains the x86 version).

A Rust compiling toolchain is required during the installation (but can be removed afterwards).

Linux

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
git clone https://github.com/neuromorphicsystems/aedat.git
cd aedat
cargo build --release
cp target/release/libaedat.so aedat.so

You can import aedat from python scripts in the same directory as aedat.so, which can be placed in any directory.

macOS

brew install rustup
rustup-init
git clone https://github.com/neuromorphicsystems/aedat.git
cd aedat
cargo build --release
cp target/release/libaedat.dylib aedat.so

You can import aedat from python scripts in the same directory as aedat.so, which can be placed in any directory.

Windows

1. install rustup (instructions availables at https://www.rust-lang.org/tools/install)
2. clone or download this repository
3. run in PowerShell from the aedat directory:

cargo build --release
copy .\target\release\aedat.dll .\aedat.pyd

You can import aedat from python scripts in the same directory as aedat.pyd, which can be placed in any directory.

Contribute

After changing any of the files in framebuffers, one must run:

flatc --rust -o src/ flatbuffers/*.fbs

To format the code, run:

cargo fmt

You may need to install rustfmt first with:

rustup component add rustfmt

Publish

1. Bump the version number in Cargo.toml.

2. Install Cubuzoa in a different directory (https://github.com/neuromorphicsystems/cubuzoa) to build pre-compiled versions for all major operating systems. Cubuzoa depends on VirtualBox (with its extension pack) and requires about 100 GB of free disk space.

cd cubuzoa
python3 -m cubuzoa provision
python3 -m cubuzoa build /path/to/aedat --post /path/to/aedat/test.py

3. Install maturin and twine

pip3 install maturin
pip3 install twine

4. Upload the compiled wheels and the source code to PyPI:

maturin sdist --out wheels
python3 -m twine upload wheels/*