nadizik/mouse-gesture-recognizer

MOUSE Gesture Recognizer (Interactive GUI Test)

A lightweight, real-time Multi-Layer Perceptron (MLP) neural network with an interactive Tkinter GUI for recognizing 2D mouse and touchpad gestures (Classes 0 to 7).

Class 0: Noise / Idle movements

Class 1–7: Target gestures

Features

• Interactive Canvas Testing: Includes a native desktop GUI (Tkinter) for drawing gestures using a mouse, touchpad, or touchscreen with real-time classification.
• Scale & Velocity Invariance: Uses 1D spatial interpolation (scipy.interpolate.interp1d) to normalize any raw drawn trajectory into exactly 64 points ($64 \times 2 = 128$ floats).
• Ultra-Compact Architecture: Lightweight MLP topology (128 -> 64 -> 32 -> 8) with LeakyReLU activations for instant inference and zero CPU overhead.
• EXE-Ready: Fully compatible with PyInstaller and multiprocessing freezes via dynamic path detection (_MEIPASS).

---

Model Architecture

Layer Index	Layer Type	Input Shape	Output Shape	Activation / Details
net.0	nn.Linear	128	64	nn.LeakyReLU(0.1)
net.2	nn.Linear	64	32	nn.LeakyReLU(0.1)
net.4	nn.Linear	32	8	Softmax (temp=0.45)

---

Quickstart & Usage

1. Download & Install Dependencies

git clone [https://huggingface.co/nadizik/mouse-gesture-recognizer](https://huggingface.co/nadizik/mouse-gesture-recognizer)
cd mouse-gesture-recognizer
pip install torch numpy scipy