Visualization READY
Each canvas shows 16×16 pixels of structured data evolving through the diffusion process. Hover over any canvas to inspect per-pixel values and noise contributions.
● Original x₀
SNR: ∞
● Noisy xₜ
SNR: –
● Reconstructed
SNR: –
Signal
Noise
0
Detailed Math Mode
CLOSED-FORM FORWARD PROCESS
xt = √ᾱt · x₀ + √(1 − ᾱt) · ε
Awaiting computation…
SIGNAL COMPONENT
√ᾱt · x₀ = –
NOISE COMPONENT
√(1−ᾱt) · ε = –
REVERSE STEP (DDPM)
xt−1 = (1/√αt) · [xt − ((1−αt)/√(1−ᾱt)) · ε̂θ(xt,t)] + σt·z
Not yet computed
PREDICTED NOISE ε̂θ
–
POSTERIOR VARIANCE σt²
β̃t = (1−ᾱt−1)/(1−ᾱt) · βt = –
SCHEDULE PARAMETERS (t = 0)
αt=–, βt=–, ᾱt=–
Noise Schedule
ᾱt (signal)
1−ᾱt (noise)
βt (step)