decodes the tensor in slices. This results in image portions that don't exactly match, so we overlap, feather, and merge to reduce

Initialize network weights. Args: module_list (list[nn.Module] | nn.Module): Modules to be initialized. scale (float): Scale init

Determine the maximum font size that allows the text to fit within the given image dimensions and margin constraints. Updated to use multilin

given a nested dictionary, print it out.

Deep network interpolation. ``Paper: Deep Network Interpolation for Continuous Imagery Effect Transition``

given a list of dotstrings, return a nested dictionary.

draw head pose axis.

draw head pose cube. Where (tdx, tdy) is the translation of the face. For pose we have [pitch yaw roll tdx tdy tdz scale_factor]

Encode the variances from the priorbox layers into the ground truth boxes we have matched (based on jaccard overlap) with the prior boxes. Arg

Encode the variances from the priorbox layers into the ground truth boxes we have matched (based on jaccard overlap) with the prior boxes. Arg

Find the likelihood of a list of textual concepts existing in the image.

Given a list of state_dict keys, collapse similar keys into patterns. For example, if the keys are: foo.bar.0.baz foo.bar.1.baz The

Flattens a nested dictionary into a dictionary with dot-separated keys. This function removes items from the original dictionary as they are

Formats the filename based on the provided template and variables.

Simplified version of multihead attention (taken from torch source code but without tons of if clauses). The mlp and layer norm come from CLI

Fourier filter as introduced in FreeU (https://arxiv.org/abs/2309.11497). This version of the method comes from here: https://github.com/Chen

Range but handles floats.

Create a dictionary of LoRA layers from a state dict. Expects the state dict to be a succession of down and up weights.

Returns bounding boxes at various zoom levels for faces in the image.

ImaginePrompt to generated image.