Create a 1D, 2D, or 3D convolution module.

Make a standard normalization layer. :param channels: number of input channels. :return: an nn.Module for normalization.

Create a linear module.

Instantiate a module from an OmegaConf-style config. This is a local copy of the utility used in ObjectStitch's ldm.util.

Zero out the parameters of a module and return it.

Imagenet Superresolution Dataloader Performs following ops in order: 1. crops a crop of size s from image either as random o

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

Create a uint8 mask (0/255) from bbox coordinates.

Pad to square using the original InsertAnything convention. Only the shorter side is padded, and padding is applied with `np.pad`.

Prepare model kwargs and conditioning vectors for sampling. Port of ObjectStitch's `prepare_input` function, adapted to our config.

Convert a BCHW tensor in [-1,1] or [0,1] to uint8 numpy images. Port of ObjectStitch's `tensor2numpy` utility.

Run SAM with a box prompt and return a binary mask (uint8 0/255). Ported from `run_sam_on_objectstitch.py::run_sam_on_image`.

:param x: img of size (bs, c, h, w) :return: n img crops of size (n, bs, c, kernel_size[0], kernel_size[1])

Create a binary mask (uint8, 0/255) from an (x1,y1,x2,y2) bbox.

A counter for the `thop` package to count the operations in an attention operation. Meant to be used like: macs, params = thop.pr

:param h: height :param w: width :return: normalized distance to image border, wtith min distance = 0 at border and

Expand a bbox using the same heuristic as the original repo. `mask` is only used for its shape; the expansion ratio is adapted based on the r

Prepare a single-sample batch for ObjectStitch. This is a slightly simplified version of `generate_image_batch` from ObjectStitch's `scripts/

In-memory version of :func:`generate_image_batch`. Parameters are identical in semantics, but accept already-loaded images.