Initialize linear weights as described in PGGAN.

Make stem layer for ResNet.

Non-Distributed training function. Args: model (nn.Module): The model to be trained. dataset (:obj:`Dataset`): Train dataset.

Calculate SSIM (structural similarity) for one channel images. It is called by func:`calculate_ssim`. Args: img1, img2 (ndarray): Im

Calculate accuracy according to the prediction and target. Args: pred (torch.Tensor | np.array): The model prediction. target (tor

Add extra files that are required to support MIM into the package. These files will be added by creating a symlink to the originals if the pa

Performs Gamma Correction on the input image. This function is adopted from skimage: https://github.com/scikit-image/scikit-image/blob/ 7

Returns the names of available CLIP models

Backward function for the discriminator. Args: outputs (dict): Dict of forward results. Returns: dict: Loss

Backward function for the discriminators. Args: outputs (dict): Dict of forward results. Returns: dict: Loss

Backward function for the generator. Args: outputs (dict): Dict of forward results. Returns: dict: Loss dict

Backward function for the generators. Args: outputs (dict): Dict of forward results. Returns: dict: Loss dic

Batched predict. Args: x (Tensor): Input tensor. coord (Tensor): coord tensor. cell (Tensor): cell tensor

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 regions.

Generate a bivariate isotropic or anisotropic Gaussian kernel. In isotropic mode, only `sig_x` is used. `sig_y` and `theta` are ignored.

Generate a bivariate generalized Gaussian kernel. Described in `Parameter Estimation For Multivariate Generalized Gaussian Distributions` by

Generate a plateau-like anisotropic kernel. This kernel has a form of 1 / (1+x^(beta)). Ref: https://stats.stackexchange.com/questions/203629

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

Build a metric calculator.

Calculate PDF of the bivariate Gaussian distribution. Args: sigma_matrix (ndarray): The variance matrix with shape (2, 2). grid (

Calculate the overlap factor after applying deconv. Args: attention_score (torch.Tensor): The attention score with shape of

Calculate (h, w) after unfolding. The official implementation of `unfold` in pytorch will put the dimension (h, w) into `L`. Thus, th

Check whether the input is a mirror-extended sequence. If mirror-extended, the i-th (i=0, ..., t-1) frame is equal to the (t-1-i)-th

Check whether the input is a mirror-extended sequence. If mirror-extended, the i-th (i=0, ..., t-1) frame is equal to the (t-1-i)-th

Check whether the input is a mirror-extended sequence. If mirror-extended, the i-th (i=0, ..., t-1) frame is equal to the (t-1-i)-th

Check whether the input is a mirror-extended sequence. If mirror-extended, the i-th (i=0, ..., t-1) frame is equal to the (t-1-i)-th

Collect the information of the running environments.

Collect results in cpu mode. It saves the results on different gpus to 'tmpdir' and collects them by the rank 0 worker. Args: re

Collect results in gpu mode. It encodes results to gpu tensors and use gpu communication for results collection. Args: result_pa

Compute features. Args: block (ndarray): 2D Image block. Returns: list: Features with length of 18.

Compute optical flow using SPyNet for feature warping. Note that if the input is an mirror-extended sequence, 'flows_forward' is not

Compute optical flow using SPyNet for feature alignment. Note that if the input is an mirror-extended sequence, 'flows_forward' is no

Compute optical flow using SPyNet for feature warping. Note that if the input is an mirror-extended sequence, 'flows_forward' is not

Compute flow from ref to supp. Note that in this function, the images are already resized to a multiple of 32. Args:

Compute guided attention score. Args: similarity_map (Tensor): Similarity map of image feature with shape (1, img

Compute keyframe features for information-refill. Since EDVR-M is used, padding is performed before feature computation. Args:

Compute similarity between image feature patches. Args: img_feat (Tensor): Image feature map of shape (1, img_c,

Compute weight with equalized learning rate. Args: module (nn.Module): A module that is wrapped with equalized lr. Retur

Compute zero padding tuple.

Convert applicable model parameters to fp16

Denormalize the output image. Args: img (Tensor): Output image. Returns: Tensor: Denormalized image.

Denormalize the output image. Args: img (Tensor): Output image. Returns: Tensor: Denormalized image.

Gradient of gaussian. Args: x (array_like): The independent variable. sigma (float): Standard deviation of the gaussian function.

On PyTorch 1.9 a CUDA fuser bug prevents the Inception JIT model to run. Refers to: https://github.com/GaParmar/clean-fid/blob/5e1e84cdea96

Evaluation function. If the output contains multiple frames, we compute the metric one by one and take an average. Args:

Evaluation function. If the output contains multiple frames, we compute the metric one by one and take an average. Args

Evaluation function. If the output contains multiple frames, we compute the metric one by one and take an average. Args

Evaluation function. Args: output (Tensor): Model output with shape (n, c, h, w). gt (Tensor): GT Tensor with shape (

Extract image feature, alpha feature unknown patches. Args: img_feat (Tensor): Image feature map of shape (N, img

Build pretrained Inception model for FID computation. The Inception model for FID computation uses a different set of weights and has a sligh

Forward function. Args: lq (Tensor): Input lq images. gt (Tensor): Ground-truth image. Default: None. tes

Defines the computation performed at every test call.

Testing forward function. Args: lq (Tensor): LQ image gt (Tensor): GT image lq_up (Tensor): Upsampled LQ

Forward function for testing. Args: img_a (Tensor): Input image from domain A. img_b (Tensor): Input image from domai

Forward function for testing. Args: img_a (Tensor): Input image from domain A. img_b (Tensor): Input image from domai

Abstract method for training forward. All subclass should overwrite it.

Defines the computation performed at every training call. Args: merged (Tensor): Image to predict alpha matte. trimap

Forward function for training. In this version, we do not use this interface.

Training forward function. This is a basic function, interpolate a frame between the given two frames. Args: inp

Training forward function. Args: lq (Tensor): LQ Tensor with shape (n, c, h, w). gt (Tensor): GT Tensor with shape (n

Forward function for training. Args: img_a (Tensor): Input image from domain A. img_b (Tensor): Input image from doma

Forward function for training. Args: img_a (Tensor): Input image from domain A. img_b (Tensor): Input image from doma

Compute the frechet distance.

Freeze the backbone and only train the refiner.

Set BatchNorm modules in the model to evaluation mode.

Fuse correlation map. This operation is to fuse correlation map for increasing large consistent correlation regions. The mec

Hard Attention. Gathers values along an axis specified by dim. Args: inputs (Tensor): The source tensor. (N, C*k*k, H*W)

Gradient of gaussian. Args: sigma (float): Standard deviation of the gaussian kernel. epsilon (float): Small value used when calc