Diffusion Model Correction Training Diagram

Overview

A two-row diagram of correction training for diffusion models. It is an editable PowerPoint diagram adapted from the original paper figure under CC BY 4.0.

What the diagram shows

Training row: a noisy image z_t → a Denoising box D_θ(z_t, t) → the prediction x̂⁽⁰⁾ → a stop-gradient (sg) mark → a Correction box C_φ(x̂, t) → the corrected x̂⁽¹⁾, with both the prediction and the corrected image feeding an L₂ loss against the target x. Sampling row: initial noise z₁ → Denoising → prediction x̂^(k) → a highlighted Correction box → corrected x̂^(k+1) → final sample, with a blue “Correction × K” loop around the correction module and a gray “Denoising step: t → s” loop back to the denoiser. Every box, arrow, label and math term is a native shape you can recolor, resize and relabel for your own slides, posters or papers.

File format

The download contains .pptx (1 slide, 11.02 × 5.58 in; opens in PowerPoint, Keynote, WPS and LibreOffice Impress; Google Slides works too, but complex freeforms may shift slightly), .svg and .drawio. Includes preview.png (full resolution) and a 1200 px web preview.

Shapes and editing

The file has 38 native editable shapes: 4 auto-shapes, 15 text boxes, 19 freeforms. Plus 8 embedded images (see below). You can edit all text and math (z_t, z₁, x̂⁽⁰⁾, x̂⁽¹⁾, x̂^(k), x̂^(k+1), D_θ, C_φ, L₂, K, t → s), the Denoising and Correction boxes, every arrow and loop, the stop-gradient mark, the Training / Sampling row labels and the divider lines.

Embedded images and fonts

The 8 example images (kiwi-and-strawberry bowl, noise and strawberry close-ups) are embedded PNG images; you can move, resize, crop or replace them, but not recolor them like vector shapes. All text and formulas use Times New Roman, which is preinstalled on Windows and macOS (LibreOffice substitutes the metric-compatible Liberation Serif). Sub- and superscripts are real formatted text runs, not equation objects. The SVG keeps text as text; the draw.io file is a shape-by-shape conversion and may need small nudges. This is a faithful redraw for reuse, not the authors' original source file.