Differentiable Spline Renderer

A Python implementation for rendering smooth, differentiable splines using PyTorch, designed for use in machine learning and generative art tasks. This repository demonstrates the creation of quadratic Bézier curves and their visualization on a canvas with customizable colors and brush sizes.

Features

Smooth spline generation: Uses quadratic Bézier interpolation to produce smooth curves.
Customizable rendering: Specify brush sizes and colors for spline rendering.
Differentiability: Built with PyTorch, enabling gradient-based optimization.
Visualization: Render and display the resulting canvas with matplotlib.

Installation

git clone https://github.com/your-username/differentiable-spline-renderer.git
cd differentiable-spline-renderer

Ensure you have PyTorch and Matplotlib installed in your environment.

Usage

1. Define Control Points

The spline requires exactly three control points for quadratic Bézier interpolation. Example:

import torch
from differentiable_spline import DifferentiableSpline

control_points = torch.tensor([
    [0.2, 0.3],
    [0.5, 0.8],
    [0.8, 0.4]
], dtype=torch.float32)

spline = DifferentiableSpline(control_points, canvas_size=(128, 128), device='cpu')

2. Interpolate and Render

Specify colors and brush sizes, then generate the canvas:

colors = torch.tensor([
    [1.0, 0.0, 0.0],  # Red
    [0.0, 1.0, 0.0],  # Green
    [0.0, 0.0, 1.0]   # Blue
])

brush_sizes = torch.tensor([10.0])  # Single brush size for all strokes

canvas = spline.generate_spline_canvas(colors, brush_sizes, num_points=300)
spline.show_canvas(canvas)

API Reference