eseg 1.1.0.post1

Models and utilities for event-based depth / segmentation (Surreal benchmark).

eseg

Event-based depth/segmentation research package with ConvLSTM models, data utilities, and live camera streaming helpers.

Features

• ConvLSTM-based models for event-stream inference
• Event voxelization and preprocessing helpers
• Utilities for HDF5/AEDAT4/RAW event data
• Live streaming pipeline for Prophesee and DAVIS cameras
• Training helpers (losses, plotting, and evaluation utilities)

Python compatibility

This release currently targets Python 3.12.

Installation

Install from PyPI:

pip install eseg

Install from source (development):

git clone <your-repository-url>
cd eseg
python -m venv .venv
source .venv/bin/activate  # Linux/macOS
pip install -e .[dev]

Optional runtime dependencies for live cameras

For camera streaming, install one or both vendor SDKs:

• Prophesee: Metavision SDK
• iniVation DAVIS: dv-processing

If you use GPU inference/training, install a CUDA-enabled PyTorch build first: https://pytorch.org/get-started/locally/

Quick start

import eseg
from eseg.models.ConvLSTM import EConvlstm

print(eseg.__version__)
model = EConvlstm(light=False)

Run live stream

python -m eseg.stream --help

Example:

python -m eseg.stream -m full --slice-time-ms 100

Development

Run tests:

pytest

License

MIT License. See LICENSE.

Notes

This is research-oriented software; interfaces may evolve between releases.