Kernel Inception Distance¶

Module Interface¶

class torchmetrics.image.kid.KernelInceptionDistance(feature=2048, subsets=100, subset_size=1000, degree=3, gamma=None, coef=1.0, reset_real_features=True, **kwargs)[source]

Calculates Kernel Inception Distance (KID) which is used to access the quality of generated images. Given by

$KID = MMD(f_{real}, f_{fake})^2$

where $MMD$ is the maximum mean discrepancy and $I_{real}, I_{fake}$ are extracted features from real and fake images, see [1] for more details. In particular, calculating the MMD requires the evaluation of a polynomial kernel function $k$

$k(x,y) = (\gamma * x^T y + coef)^{degree}$

which controls the distance between two features. In practise the MMD is calculated over a number of subsets to be able to both get the mean and standard deviation of KID.

Using the default feature extraction (Inception v3 using the original weights from [2]), the input is expected to be mini-batches of 3-channel RGB images of shape (3 x H x W) with dtype uint8. All images will be resized to 299 x 299 which is the size of the original training data.

Note

using this metric with the default feature extractor requires that torch-fidelity is installed. Either install as pip install torchmetrics[image] or pip install torch-fidelity

Parameters

feature¶ (Union[str, int, Module]) –
Either an str, integer or nn.Module:
- an str or integer will indicate the inceptionv3 feature layer to choose. Can be one of the following: ‘logits_unbiased’, 64, 192, 768, 2048
- an nn.Module for using a custom feature extractor. Expects that its forward method returns an [N,d] matrix where N is the batch size and d is the feature size.
subsets¶ (int) – Number of subsets to calculate the mean and standard deviation scores over
subset_size¶ (int) – Number of randomly picked samples in each subset
degree¶ (int) – Degree of the polynomial kernel function
gamma¶ (Optional[float]) – Scale-length of polynomial kernel. If set to None will be automatically set to the feature size
coef¶ (float) – Bias term in the polynomial kernel.
reset_real_features¶ (bool) – Whether to also reset the real features. Since in many cases the real dataset does not change, the features can cached them to avoid recomputing them which is costly. Set this to False if your dataset does not change.
kwargs¶ (Dict[str, Any]) – Additional keyword arguments, see Advanced metric settings for more info.

References

[1] Demystifying MMD GANs Mikołaj Bińkowski, Danica J. Sutherland, Michael Arbel, Arthur Gretton https://arxiv.org/abs/1801.01401

[2] GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, Sepp Hochreiter https://arxiv.org/abs/1706.08500

Raises

ValueError – If feature is set to an int (default settings) and torch-fidelity is not installed
ValueError – If feature is set to an int not in [64, 192, 768, 2048]
ValueError – If subsets is not an integer larger than 0
ValueError – If subset_size is not an integer larger than 0
ValueError – If degree is not an integer larger than 0
ValueError – If gamma is niether None or a float larger than 0
ValueError – If coef is not an float larger than 0
ValueError – If reset_real_features is not an bool

Example

>>> import torch
>>> _ = torch.manual_seed(123)
>>> from torchmetrics.image.kid import KernelInceptionDistance
>>> kid = KernelInceptionDistance(subset_size=50)
>>> # generate two slightly overlapping image intensity distributions
>>> imgs_dist1 = torch.randint(0, 200, (100, 3, 299, 299), dtype=torch.uint8)
>>> imgs_dist2 = torch.randint(100, 255, (100, 3, 299, 299), dtype=torch.uint8)
>>> kid.update(imgs_dist1, real=True)
>>> kid.update(imgs_dist2, real=False)
>>> kid_mean, kid_std = kid.compute()
>>> print((kid_mean, kid_std))
(tensor(0.0337), tensor(0.0023))

Initializes internal Module state, shared by both nn.Module and ScriptModule.

compute()[source]

Calculate KID score based on accumulated extracted features from the two distributions. Returns a tuple of mean and standard deviation of KID scores calculated on subsets of extracted features.

Implementation inspired by Fid Score

Return type: Tuple[Tensor, Tensor]

reset()[source]

This method automatically resets the metric state variables to their default value.

Return type: None

update(imgs, real)[source]

Update the state with extracted features.

Parameters

imgs¶ (Tensor) – tensor with images feed to the feature extractor
real¶ (bool) – bool indicating if imgs belong to the real or the fake distribution

Return type

None