Clean image

As explained in Build a Text-to-Image Generator (from Scratch)

A clean image is an image containing the original visual signal without diffusion noise.

A clean image contains the original visual signal without diffusion noise. It is represented as x0, the time-step-zero starting point for forward diffusion, before noise produces progressively noisier images. section 1.3.1 section 5.1.1

Forward diffusion uses the clean image as the first stage in a controlled sequence:

• Start with x0 at time step 0. section 1.3.1
• Add small amounts of normally distributed noise at successive time steps, forming weighted combinations of the image and noise. section 5.1.1
• Continue until the image at T = 1,000 approximates pure random noise. section 1.3.1
• Use reverse diffusion to reconstruct a clean image from noisy inputs by repeatedly predicting and removing noise. section 5.2

Visualizing the signal being lost

A dog image serves as x0: it appears clean at time step 0, becomes partially obscured as noise increases, and ends as random noise. The same progression can be plotted for clean flower images at steps 0, 200, 400, 600, 800, and 1,000. section 1.3.1 section 7.2.2

Figure 1.6 In the forward diffusion process, we start with a clean image from the training set, x₀, and add noise ε₀ to it to form a noisy image x₁, which is a weighted sum of x₀ and ε₀. We repeat this process for 1,000 time steps until the image x_1,000 becomes random noise.

Trace the clean image x0 through 1,000 forward-diffusion steps until it becomes random noise. section 1.3.1

How Build a Text-to-Image Generator (from Scratch) covers it

How deep it goes

Defines x0, explains weighted noise addition and time steps, and stops short of detailed denoising-model construction.

Examples

Uses dog images, clean flower training images, DDIM schedules, and plots across progressively noisier time steps.

What it assumes

Assumes familiarity with images and random noise while introducing diffusion through discrete time steps and weighted combinations.

What you can do after

Apply it via a DDIM noise scheduler to create and visualize noisy training images from clean inputs.

How it connects

Connects forward diffusion with noisy images, time steps, reverse diffusion, and denoising models.

liveBookmanning.com

Read Build a Text-to-Image Generator (from Scratch) on liveBook Get the book at manning.com

Related concepts

• denoisingproduced by · Denoising is the process of removing unwanted noise or corruption from data while preserving meaningful structure.
• denoising autoencoderused by · A denoising autoencoder is an autoencoder trained to reconstruct clean data from deliberately corrupted or noisy inputs.
• noisecontrasts with · Noise is variation, content, or data treated as unwanted, irrelevant, or distinct from the signal, pattern, or structure of interest.
• data cleaningproduced by · Data cleaning is the process of detecting and correcting, removing, transforming, or otherwise handling inaccurate, incomplete, inconsistent, duplicate, or unsuitable data.
• image generationcontrasts with · Image generation is the computational creation of new images from learned models, input data, or specified conditions.

This explainer was generated by Manning's knowledge graph from the text of Build a Text-to-Image Generator (from Scratch). The section links point to the parts of the book it draws on.