1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267
|
import warnings
from typing import Callable, Optional, Sequence, Union
import torch
from packaging.version import Version
from ignite.metrics.gan.utils import _BaseInceptionMetric, InceptionModel
from ignite.metrics.metric import reinit__is_reduced, sync_all_reduce
__all__ = [
"FID",
]
if Version(torch.__version__) <= Version("1.7.0"):
torch_outer = torch.ger
else:
torch_outer = torch.outer
def fid_score(
mu1: torch.Tensor, mu2: torch.Tensor, sigma1: torch.Tensor, sigma2: torch.Tensor, eps: float = 1e-6
) -> float:
try:
import numpy as np
except ImportError:
raise ModuleNotFoundError("fid_score requires numpy to be installed.")
try:
import scipy.linalg
except ImportError:
raise ModuleNotFoundError("fid_score requires scipy to be installed.")
mu1, mu2 = mu1.cpu(), mu2.cpu()
sigma1, sigma2 = sigma1.cpu(), sigma2.cpu()
diff = mu1 - mu2
# Product might be almost singular
covmean, _ = scipy.linalg.sqrtm(sigma1.mm(sigma2), disp=False)
# Numerical error might give slight imaginary component
if np.iscomplexobj(covmean):
if not np.allclose(np.diagonal(covmean).imag, 0, atol=1e-3):
m = np.max(np.abs(covmean.imag))
raise ValueError("Imaginary component {}".format(m))
covmean = covmean.real
tr_covmean = np.trace(covmean)
if not np.isfinite(covmean).all():
tr_covmean = np.sum(np.sqrt(((np.diag(sigma1) * eps) * (np.diag(sigma2) * eps)) / (eps * eps)))
return float(diff.dot(diff).item() + torch.trace(sigma1) + torch.trace(sigma2) - 2 * tr_covmean)
class FID(_BaseInceptionMetric):
r"""Calculates Frechet Inception Distance.
.. math::
\text{FID} = |\mu_{1} - \mu_{2}| + \text{Tr}(\sigma_{1} + \sigma_{2} - {2}\sqrt{\sigma_1*\sigma_2})
where :math:`\mu_1` and :math:`\sigma_1` refer to the mean and covariance of the train data and
:math:`\mu_2` and :math:`\sigma_2` refer to the mean and covariance of the test data.
More details can be found in `Heusel et al. 2017`__
__ https://arxiv.org/pdf/1706.08500.pdf
In addition, a faster and online computation approach can be found in `Mathiasen et al. 2020`__
__ https://arxiv.org/pdf/2009.14075.pdf
Remark:
This implementation is inspired by `pytorch_fid` package which can be found `here`__
__ https://github.com/mseitzer/pytorch-fid
.. note::
The default Inception model requires the `torchvision` module to be installed.
FID also requires `scipy` library for matrix square root calculations.
Args:
num_features: number of features predicted by the model or the reduced feature vector of the image.
Default value is 1000.
feature_extractor: a torch Module for extracting the features from the input data.
It returns a tensor of shape (batch_size, num_features).
If neither ``num_features`` nor ``feature_extractor`` are defined, by default we use an ImageNet
pretrained Inception Model and use model's output logits as features.
If only ``num_features`` is defined but ``feature_extractor`` is not defined,
``feature_extractor`` is assigned Identity Function.
Please note that the model will be implicitly converted to device mentioned in the ``device``
argument.
output_transform: a callable that is used to transform the
:class:`~ignite.engine.engine.Engine`'s ``process_function``'s output into the
form expected by the metric. This can be useful if, for example, you have a multi-output model and
you want to compute the metric with respect to one of the outputs.
By default, metrics require the output as ``(y_pred, y)`` or ``{'y_pred': y_pred, 'y': y}``.
device: specifies which device updates are accumulated on. Setting the
metric's device to be the same as your ``update`` arguments ensures the ``update`` method is
non-blocking. By default, CPU.
Examples:
For more information on how metric works with :class:`~ignite.engine.engine.Engine`, visit :ref:`attach-engine`.
.. include:: defaults.rst
:start-after: :orphan:
.. testcode::
metric = FID(num_features=1, feature_extractor=default_model)
metric.attach(default_evaluator, "fid")
y_true = torch.ones(10, 4)
y_pred = torch.ones(10, 4)
state = default_evaluator.run([[y_pred, y_true]])
print(state.metrics["fid"])
.. testoutput::
0.0
.. note::
The default `torchvision` model used is InceptionV3 pretrained on ImageNet.
This can lead to differences in results with `pytorch_fid`. To find comparable results,
the following model wrapper should be used:
.. code::
import torch.nn as nn
# wrapper class as feature_extractor
class WrapperInceptionV3(nn.Module):
def __init__(self, fid_incv3):
super().__init__()
self.fid_incv3 = fid_incv3
@torch.no_grad()
def forward(self, x):
y = self.fid_incv3(x)
y = y[0]
y = y[:, :, 0, 0]
return y
# use cpu rather than cuda to get comparable results
device = "cpu"
# pytorch_fid model
dims = 2048
block_idx = InceptionV3.BLOCK_INDEX_BY_DIM[dims]
model = InceptionV3([block_idx]).to(device)
# wrapper model to pytorch_fid model
wrapper_model = WrapperInceptionV3(model)
wrapper_model.eval();
# comparable metric
pytorch_fid_metric = FID(num_features=dims, feature_extractor=wrapper_model)
Important, `pytorch_fid` results depend on the batch size if the device is `cuda`.
.. versionadded:: 0.4.6
"""
_state_dict_all_req_keys = ("_num_examples", "_train_total", "_test_total", "_train_sigma", "_test_sigma")
def __init__(
self,
num_features: Optional[int] = None,
feature_extractor: Optional[torch.nn.Module] = None,
output_transform: Callable = lambda x: x,
device: Union[str, torch.device] = torch.device("cpu"),
) -> None:
try:
import numpy as np # noqa: F401
except ImportError:
raise ModuleNotFoundError("This module requires numpy to be installed.")
try:
import scipy # noqa: F401
except ImportError:
raise ModuleNotFoundError("This module requires scipy to be installed.")
if num_features is None and feature_extractor is None:
num_features = 1000
feature_extractor = InceptionModel(return_features=False, device=device)
self._eps = 1e-6
super(FID, self).__init__(
num_features=num_features,
feature_extractor=feature_extractor,
output_transform=output_transform,
device=device,
)
@staticmethod
def _online_update(features: torch.Tensor, total: torch.Tensor, sigma: torch.Tensor) -> None:
total += features
sigma += torch_outer(features, features)
def _get_covariance(self, sigma: torch.Tensor, total: torch.Tensor) -> torch.Tensor:
r"""
Calculates covariance from mean and sum of products of variables
"""
sub_matrix = torch_outer(total, total)
sub_matrix = sub_matrix / self._num_examples
return (sigma - sub_matrix) / (self._num_examples - 1)
@reinit__is_reduced
def reset(self) -> None:
self._train_sigma = torch.zeros(
(self._num_features, self._num_features), dtype=torch.float64, device=self._device
)
self._train_total = torch.zeros(self._num_features, dtype=torch.float64, device=self._device)
self._test_sigma = torch.zeros(
(self._num_features, self._num_features), dtype=torch.float64, device=self._device
)
self._test_total = torch.zeros(self._num_features, dtype=torch.float64, device=self._device)
self._num_examples: int = 0
super(FID, self).reset() # type: ignore
@reinit__is_reduced
def update(self, output: Sequence[torch.Tensor]) -> None:
train, test = output
train_features = self._extract_features(train)
test_features = self._extract_features(test)
if train_features.shape[0] != test_features.shape[0] or train_features.shape[1] != test_features.shape[1]:
raise ValueError(
f"""
Number of Training Features and Testing Features should be equal ({train_features.shape} != {test_features.shape})
"""
)
# Updates the mean and covariance for the train features
for features in train_features:
self._online_update(features, self._train_total, self._train_sigma)
# Updates the mean and covariance for the test features
for features in test_features:
self._online_update(features, self._test_total, self._test_sigma)
self._num_examples += train_features.shape[0]
@sync_all_reduce("_num_examples", "_train_total", "_test_total", "_train_sigma", "_test_sigma")
def compute(self) -> float:
fid = fid_score(
mu1=self._train_total / self._num_examples,
mu2=self._test_total / self._num_examples,
sigma1=self._get_covariance(self._train_sigma, self._train_total),
sigma2=self._get_covariance(self._test_sigma, self._test_total),
eps=self._eps,
)
if torch.isnan(torch.tensor(fid)) or torch.isinf(torch.tensor(fid)):
warnings.warn("The product of covariance of train and test features is out of bounds.")
return fid
|