Uni-LaDiR blog figures

• fig1.pdf: original manuscript Figure 1, fig1-0912-motion-detail-v76.pdf.
• fig2.pdf: original manuscript Figure 2, method-sequence-v23.pdf.
• render.py: renders the original vectors to SVG/PNG and replots measured results using Matplotlib. Run with Python, PyMuPDF, and Matplotlib installed.
• results.json: scores used by the plots. Main panels use manuscript Tables 1 and 3; sharing panels use Figure 3 and supplementary sharing tables. All are percentages. Relative gains use the baseline denominator.

Source manuscript supplied by the author on September 14, 2026. The figures’ teacher paths are training views; they do not imply access to future teacher observations at inference. Scene images illustrate the concept. The method walkthrough highlights existing diagram regions and is not a simulation of learned diffusion trajectories.

Desktop result charts show all comparisons side by side. Mobile charts reflow the same values as horizontal bars. All axes start at zero; no uncertainty has been invented. The paper distinguishes matched experiments from reported baseline comparisons; see the blog captions.