Merge pull request #54 from Qbiwan/change_xt

change x_0 to x_t

Author: johnowhitaker

unit4/01_ddim_inversion.ipynb

@@ -242,10 +242,11 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "During sampling, we begin with pure noise at timestep 1000 and slowly move towards timestep 0. To calculate the next t in the sampling trajectory ($x_{t-1}$ since we're moving from high t to low t) we predict the noise ($\\epsilon_\\theta(x_t)$, which is the output of our model) and use this to calculate the predicted denoised image $x_0$. Then we use this prediction to move a small distance in the 'direction pointing to $x_0$'. Finally, we can add some additional noise scaled by $\\sigma_t$. Here's the relevant section from the paper showing this in action:"
+    "During sampling, we begin with pure noise at timestep 1000 and slowly move towards timestep 0. To calculate the next t in the sampling trajectory ($x_{t-1}$ since we're moving from high t to low t) we predict the noise ($\\epsilon_\\theta(x_t)$, which is the output of our model) and use this to calculate the predicted denoised image $x_0$. Then we use this prediction to move a small distance in the 'direction pointing to $x_t$'. Finally, we can add some additional noise scaled by $\\sigma_t$. Here's the relevant section from the paper showing this in action:"
    ]
   },
   {