Generate mask in ndarray from bbox. The returned mask has the shape of (h, w, 1). '1' indicates the hole and '0' indicates the valid regio

finds a function handle with the name given as 'name' in config, and returns the instance initialized with corresponding args.

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

set parameter to gpu or ddp

must use set_device in tensor

Zero out the parameters of a module and return it.

Generate free-form mask. The method of generating free-form mask is in the following paper: Free-Form Image Inpainting with Gated Convolut

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

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

Generate a random bbox for the mask on a given image. In our implementation, the max value cannot be obtained since we use `np.random.rand

Generate random irregular masks. This is a modified version of free-form mask implemented in 'brush_stroke_mask'. We prefer to use

save network structure, only work on GPU 0

convert AB and L to RGB

create_model

create train/test dataloader and validation dataloader, validation dataloader is None when phase is test or not GPU 0

loading Dataset() class from given file's name

define network with weights initialization

convert to NoneDict, which return None for missing key.

Create sinusoidal timestep embeddings. :param gammas: a 1-D Tensor of N indices, one per batch element. These may be fr

Get irregular mask with the constraints in mask ratio Args: img_shape (tuple[int]): Size of the image. area_ratio_range (tupl

Computes the inception score of the generated images imgs imgs -- Torch dataset of (3xHxW) numpy images normalized in the range [-1, 1] cuda

initialize network's weights init_type: normal | xavier | kaiming | orthogonal https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/bl

save pretrained model and training state, which only do on GPU 0.

threads running on each GPU

resume the optimizers and schedulers for training, only work when phase is test or resume training enable

saves training state during training, only work on GPU 0

set up logger

split a dataset into non-overlapping new datasets of given lengths. main code is from random_split function in pytorch

Converts a torch Tensor into an image Numpy array Input: 4D(B,(3/1),H,W), 3D(C,H,W), or 2D(H,W), any range, RGB channel order Output: 3D(

If visualization is configured to use: return add_data() methods of tensorboard with additional information (step, tag) added.