tcasl 1.0.3

Python library for temporal contrast American Sign Language classification.

TCASLCore

A lightweight Python inference engine for real-time American Sign Language (ASL) recognition using temporal contrast emulation to simulate a Dynamic Vision Sensor (DVS).

Features

• Temporal contrast emulation: Converts standard webcam frames into sparse DVS-style event maps, isolating hand motion and discarding static background.
• Auto-downloading weights: Model weights are fetched automatically from GitHub Releases and cached locally via torch.hub, no manual download required.
• Auto-formatting: Center-crops and rescales raw frames to the 128×128 input resolution expected by the network.
• Multiple model versions: Switch between sdnn_v1 and sdnn_v2 via a single argument, or supply your own local .pth file.

Installation

pip install tcasl

Quick Start

Full examples (static image and real-time webcam) are in examples/.

Available Models

Tag	Description
sdnn_v2 (default)	Latest release (97.3% Accuracy)
sdnn_v1	Initial release (95.9% Accuracy)

API Reference

Full docstrings are in src/tcasl/core.py.

TCASL Project

For background and the research paper, see the main TCASL repository.