This post shares my open-source repository Visualize-Your-SDEs, with a specific focus on the Diffusion Bridge demo:
- GitHub: Bili-Sakura/Visualize-Your-SDEs
- Live site: bili-sakura.github.io/Visualize-Your-SDEs
Diffusion Bridge Demo (Embedded)
The embedded app below is the diffusion-bridge page directly from the project:
If your browser blocks the embed, open it directly here:
Open Diffusion Bridge Demo
What you can control
This diffusion-bridge visualization is designed for fast experimentation:
- Model type: DDBM, I2SB, DDIB, and Turbo/CUT style direct mapping.
- Path count and step count: control simulation granularity and visual complexity.
- Noise level: adjust bridge stochasticity with a sigma slider.
- Source/target distributions: single, bimodal, trimodal, uniform, and gaussian options.
- Appearance options: color maps, backgrounds, legend and heatmap toggles.
- Export: PNG/SVG/PDF-style export from the rendered Plotly figure.
Why this is useful
A diffusion bridge is an intuitive way to study conditional transport from a source distribution to a target distribution over time. In this demo you can inspect:
- Trajectories of sampled paths.
- Density evolution as a heatmap.
- Endpoint marginals on source and target sides.
Together these views make it easier to build intuition for how bridge schedules and noise profiles affect generated dynamics.