torcheval.metrics.FrechetInceptionDistance

class torcheval.metrics.FrechetInceptionDistance(model: Optional[Module] = None, feature_dim: int = 2048, device: Optional[device] = None)

__init__(model: Optional[Module] = None, feature_dim: int = 2048, device: Optional[device] = None) → None

Computes the Frechet Inception Distance (FID) between two distributions of images (real and generated).

The original paper: https://arxiv.org/pdf/1706.08500.pdf

Parameters:

• model (nn.Module) – Module used to compute feature activations. If None, a default InceptionV3 model will be used.
• feature_dim (int) – The number of features in the model’s output, the default number is 2048 for default InceptionV3.
• device (torch.device) – The device where the computations will be performed. If None, the default device will be used.

Methods

__init__([model, feature_dim, device])	Computes the Frechet Inception Distance (FID) between two distributions of images (real and generated).
compute()	Compute the FID.
load_state_dict(state_dict[, strict])	Loads metric state variables from state_dict.
merge_state(metrics)	Merge the state of another FID instance into this instance.
reset()	Reset the metric state variables to their default value.
state_dict()	Save metric state variables in state_dict.
to(device, *args, **kwargs)	Move tensors in metric state variables to device.
update(images, is_real)	Update the states with a batch of real and fake images.

Attributes

device

The last input device of Metric.to().