Build pretrained InceptionV3 Parameters ---------- output_blocks : list of int Indices of blocks to return featur

Args: dim_lst: channel dimensions. [int] alpha: channel width factor. float.

In the backward pass we receive a Tensor containing the gradient of the loss with respect to the output, and we need to compute the g

Calculates the FID of two paths Args: paths: tuple/list. len = 2. paths[0]: path for generated images. path[1]: path

Convert a batch of images (N,W,H,C) to a single big image (W*n, H*m, C) Input: batch: type=ndarray, shape=(N,W,H,C) Return: rows: type=ndarray

Calculate perceptual loss based on two vgg features. Agrs: vgg_features1, vgg_features2: vgg features. Output: loss_pe

Calculation of the statistics used by the FID. Params: -- files : List of image files paths -- model : Instance of inception m

Numpy implementation of the Frechet Distance. The Frechet distance between two multivariate Gaussians X_1 ~ N(mu_1, C_1) and X_2 ~ N(mu_2, C_2

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

Calculates the activations of the pool_3 layer for all images. Params: -- files : List of image files paths -- model : Instan

Find parameter numbers in the model. in Million.

pytorch soft-sign function

Construct a dataset with all images from a dir. dataset_dir: str. img folder path

Construct a dataset with all images from a dir. dataset: str. dataset name style: str. 'A2B' or 'B2A'

Get Inception feature maps Parameters ---------- inp : torch.autograd.Variable Input tensor of shape Bx3xHxW. Val

In the forward pass we apply the quantizer