Get a term for the variational lower-bound. The resulting units are bits (rather than nats, as one might expect). This allow

Returns a scale-and-shift version of pred according to gt Args: pred: predicted image (b, c, h, w) gt : ground truth image

Create a 1D, 2D, or 3D average pooling module.

Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [

Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [

Evaluate a function without caching intermediate activations, allowing for reduced memory at the expense of extra compute in the backward pas

Compute the ratio between d_alpha and alpha

Compute the diffusion term of the SDE Args: x: [batch_dim, ...], data point t: [batch_dim,], time vector form: s

Compute the mean for the previous step, given a function cond_fn that computes the gradient of a conditional log probability with res

Convert fm trajectory to dm trajectory using the fm t We use linear scaling here

Returns log Signal-to-Noise ratio for time step t and image size 64 eps: avoid division by zero

Sample x_{t-1} from the model using DDIM. Same usage as p_sample().

Use DDIM to sample from the model and yield intermediate samples from each timestep of DDIM. Same usage as p_sample_loop_prog

Args: z: torch.Tensor, shape (b, latent_channels, h, w)

Args: z: latent code tensor (B, C, H, W)

Compute the log-likelihood of a Gaussian distribution discretizing to a given image. :param x: the target images. It is assumed that thi

Compute the log-likelihood of a Gaussian distribution discretizing to a given image. :param x: the target images. It is assumed that this

Convert distance to focal point to planar depth. Adapted from https://github.com/apple/ml-hypersim/issues/9#issuecomment-754935697 Args:

Args: x: torch.Tensor, shape (b, 3, h, w) in [-1, 1]

Args: x: input tensor (B, C, H, W) in range [-1, 1] scaled with self.scale and shifted with self.shift

Fix for linear beta schedules to ensure that the SNR at the end of the diffusion chain is 0 (zero terminal SNR), according to the paper

Returns the scale and shift between two images More precisely, we find (b1, b2) such that img1 = b1 + b2*img2 We mod

This is the deprecated API for creating beta schedules. See get_named_beta_schedule() for the new library of schedules.

Returns the scale and shift between two images More precisely, we find (b1, b2) such that img1 = b1 + b2*img2 We mod

Wrapper function: transfrom velocity prediction model to score Args: velocity: [batch_dim, ...] shaped tensor; velocity model outp

v-parameterization: sqrt(alpha_bar_t) * eps - sqrt(1 - alpha_bar_t) * x_0

eps is the SD eps-parameterized vector field with the FM vector field is defined as z - eps First of all convert the x_t fro

v is the SD v-parameterized vector field with v = sqrt(alpha_cumprod) * eps - sqrt(1 - alpha_cumprod) * z the FM vector field is defi

guess latent channel count based on encoder filename

Args: ims: Tensor of shape (b, c, h, w) stack: "row" or "col" split: If 'row' stack by rows, if 'col' stack by columns.

Interpolate nans in a depth image by using the closest non-nan value.

Get the last step function of the SDE solver

Sample x_{t-1} from the model at the given timestep.

Sample x_{t-1} from the model at the given timestep. :param model: the model to sample from. :param x: the current tensor at

Generate samples from the model.

Generate samples from the model and yield intermediate samples from each timestep of diffusion. Arguments are the same as p_s

Normalize each sample in a batch independently with min-max normalization to [0, 1]

Args: depth: depth map in numpy format and range [0, inf] w. shape (H, W) dataset_name: name of the dataset (e.g. vkitti2, hypers

get dataset name

Get the prior KL term for the variational lower-bound, measured in bits-per-dim. This term can't be optimized, as it only depends on

Get the distribution q(x_t | x_0).

Get the distribution q(x_t | x_0). :param x_start: the [N x C x ...] tensor of noiseless inputs. :param t: the number of diff

Diffuse the data for a given number of diffusion steps. In other words sample from q(x_t | x_0). Args: x_start: