tensordot(x, y, 1).

Fused `tf.nn.conv2d()` followed by `downsample_2d()`. Padding is performed only once at the beginning, not between the operations. The fu

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

Return a tensor of size (TODO).

Calculates the activations of the pool_3 layer for all images. Params: -- images : Images -- model : Instance of inception mod

Create a function to give the output of the score-based model. Args: model: The score model. train: `True` for training and `False` f

Create a loss function for training with rectified flow. Args: sde: An `sde_lib.SDE` object that represents the forward SDE. train: `

Get rectified flow sampler Returns: A sampling function that returns samples and the number of function evaluations during sampling.

Copied and modified from https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/nn.py#L90

Get spectral norm of Dg^2 the hessian of g :param y: :param x_hat: :param sigma: :param interpolation:

r"""Fills the input `Tensor` with values according to the method desrcibed in `Delving deep into rectifiers: Surpassing human-level performanc

Optimizes with warmup and gradient clipping (disabled if negative).

Power iteration algorithm for spectral norm calculation

r""" Soft thresholding of the wavelet coefficients. :param torch.Tensor x: wavelet coefficients. :param float, torch.Tensor t

Running the inception network. Assuming input is within [0, 255].

Fused `upsample_2d()` followed by `tf.nn.conv2d()`. Padding is performed only once at the beginning, not between the operations.

Ported from JAX.

Return: sol: solution (j, err_history): j is the number of iterations used to achieve the target accuracy;

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