Spaces:
Sleeping
Sleeping
File size: 29,012 Bytes
b20c769 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 |
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchmetrics import Accuracy, F1Score, JaccardIndex, Metric
from torchmetrics.segmentation import MeanIoU
class MetricsAccuracy(Metric):
"""
Computes the Ovearall Accuracy
Args:
modalities (list): list of modalities used
num_classes (int): number of classes
save_results (bool): if True saves prediction in a csv file
get_classes (bool): if True returns the classwise F1 Score
"""
def __init__(
self,
modalities: list = [],
num_classes: int = 15,
save_results: bool = False,
get_classes: bool = False,
multilabel: bool = False,
):
super().__init__()
self.get_classes = get_classes
task = "multilabel" if multilabel else "multiclass"
self.acc = Accuracy(task=task, num_classes=num_classes)
self.save_results = save_results
self.multiclass = not (multilabel)
if save_results:
self.results = {}
def update(self, pred, gt):
self.acc(pred, gt["label"])
if self.save_results:
for i, name in enumerate(gt["name"]):
self.results[name] = list(pred.cpu()[i].numpy())
def compute(self):
return {"OA": self.acc.compute()}
class MetricsMonoModal(Metric):
"""
Computes the micro, macro and weighted F1 Score for multi label classification
Args:
modalities (list): list of modalities used
num_classes (int): number of classes
save_results (bool): if True saves prediction in a csv file
get_classes (bool): if True returns the classwise F1 Score
"""
def __init__(
self,
modalities: list = [],
num_classes: int = 15,
save_results: bool = False,
get_classes: bool = False,
multilabel: bool = True,
):
super().__init__()
self.get_classes = get_classes
task = "multilabel" if multilabel else "multiclass"
self.f1 = F1Score(
task=task, average="none", num_labels=num_classes, num_classes=num_classes
)
self.f1_micro = F1Score(
task=task, average="micro", num_labels=num_classes, num_classes=num_classes
)
self.f1_weighted = F1Score(
task=task, average="weighted", num_labels=num_classes, num_classes=num_classes
)
self.save_results = save_results
self.multiclass = not (multilabel)
if save_results:
self.results = {}
def update(self, pred, gt):
if self.multiclass:
gt["label"] = gt["label"].argmax(dim=1)
self.f1(pred, gt["label"])
self.f1_micro(pred, gt["label"])
self.f1_weighted(pred, gt["label"])
if self.save_results:
for i, name in enumerate(gt["name"]):
self.results[name] = list(pred.cpu()[i].numpy())
def compute(self):
if self.get_classes:
f1 = self.f1.compute()
out = {
"F1_Score_macro": sum(f1) / len(f1),
"F1_Score_micro": self.f1_micro.compute(),
"F1_Score_weighted": self.f1_weighted.compute(),
}
for i in range(len(f1)):
out["_".join(["F1_classe", str(i)])] = f1[i]
return out
f1 = self.f1.compute()
out = {
"F1_Score_macro": sum(f1) / len(f1),
"F1_Score_micro": self.f1_micro.compute(),
"F1_Score_weighted": self.f1_weighted.compute(),
}
if self.save_results:
out["results"] = self.results
return out
return out
class NoMetrics(Metric):
"""
Computes no metrics or saves a batch of reconstruction to visualise them
Args:
save_reconstructs (bool): if True saves a batch of reconstructions
modalities (list): list of modalities used
save_dir (str): where to save reconstructions
"""
def __init__(
self,
save_reconstructs: bool = False,
modalities: list = [],
save_dir: str = "",
):
super().__init__()
self.save_dir = save_dir
self.save_recons = save_reconstructs
self.modalities = modalities
if self.save_recons:
if not os.path.exists(save_dir):
os.makedirs(save_dir)
self.saves = {}
for modality in self.modalities:
self.saves[modality] = []
self.saves["_".join(["gt", modality])] = []
def update(self, pred, gt):
if self.save_recons:
recons, _ = pred
for modality in self.modalities:
if modality == "aerial":
preds = recons["_".join(["reconstruct", modality])]
target = gt[modality][:, :, :300, :300]
else:
preds, mask = recons["_".join(["reconstruct", modality])]
target = gt[modality][mask[:, 0], mask[:, 1]]
indice = torch.randint(0, len(preds), (1,)).item()
self.saves[modality].append(preds[indice])
self.saves["_".join(["gt", modality])].append(target[indice])
def compute(self):
if self.save_recons:
for key in self.saves.keys():
for i, tensor in enumerate(self.saves[key]):
torch.save(tensor.cpu(), self.save_dir + key + str(i) + ".pt")
return {}
class MetricsContrastif(Metric):
"""
Computes metrics for contrastive. Given embeddings for all tokens, we compute the cosine similarity matrix.
The metric computed is the accuracy of the M -1 minimum distances of each line (except diagonal of course)
being the same token across other modalities with M the number of modalities.
Args:
modalities (list): list of modalities used
"""
def __init__(
self,
modalities: list = [],
):
super().__init__()
self.modalities = modalities
self.n_k = len(self.modalities)
self.add_state("count", default=torch.tensor(0), dist_reduce_fx="sum")
for i in range(len(modalities)):
self.add_state(modalities[i], default=torch.tensor(0.0), dist_reduce_fx="sum")
def update(self, logits):
size = len(logits) // self.n_k
labels = torch.arange(size).unsqueeze(1)
labels = torch.cat([labels + i * len(labels) for i in range(self.n_k)], dim=1)
labels = torch.cat([labels for _ in range(self.n_k)]).to(logits.device)
for i in range(self.n_k):
_, top_indices = torch.topk(
logits[i * size : (i + 1) * size], k=self.n_k, dim=1, largest=True
)
self.__dict__[self.modalities[i]] += (
torch.sum(
torch.tensor(
[
top_indices[i, j] in labels[i]
for i in range(top_indices.size(0))
for j in range(self.n_k)
]
)
)
- len(top_indices)
) / (self.n_k - 1)
self.count += len(logits)
def compute(self):
dict = {}
for i in range(len(self.modalities)):
dict["_".join(["acc", self.modalities[i]])] = (
self.__dict__[self.modalities[i]] / self.count
)
return dict
class MetricsContrastifMulti(Metric):
"""
Computes metrics for contrastive. Given embeddings for all tokens, we compute the cosine similarity matrix.
The metric computed is the accuracy of the M -1 minimum distances of each line (except diagonal of course)
being the same token across other modalities with M the number of modalities.
Args:
modalities (list): list of modalities used
"""
def __init__(
self,
modalities: dict = {},
):
super().__init__()
self.modalities = modalities
for dataset in self.modalities.keys():
self.add_state(dataset + "_count", default=torch.tensor(0), dist_reduce_fx="sum")
for i in range(len(modalities[dataset])):
self.add_state(
dataset + "_" + modalities[dataset][i],
default=torch.tensor(0.0),
dist_reduce_fx="sum",
)
def update(self, logits, dataset):
modalities = self.modalities[dataset]
n_modalities = len(modalities)
size = len(logits) // n_modalities
labels = torch.arange(size).unsqueeze(1)
labels = torch.cat([labels + i * len(labels) for i in range(n_modalities)], dim=1)
labels = torch.cat([labels for _ in range(n_modalities)]).to(logits.device)
for i in range(n_modalities):
_, top_indices = torch.topk(
logits[i * size : (i + 1) * size], k=n_modalities, dim=1, largest=True
)
self.__dict__[dataset + "_" + modalities[i]] += (
torch.sum(
torch.tensor(
[
top_indices[i, j] in labels[i]
for i in range(top_indices.size(0))
for j in range(n_modalities)
]
)
)
- len(top_indices)
) / (n_modalities - 1)
self.__dict__[dataset + "_count"] += len(logits)
def compute(self):
dict = {}
for dataset in self.modalities.keys():
for i in range(len(self.modalities[dataset])):
dict["_".join(["acc", dataset, self.modalities[dataset][i]])] = (
self.__dict__[dataset + "_" + self.modalities[dataset][i]]
/ self.__dict__[dataset + "_count"]
)
return dict
class MetricsSemSeg(Metric):
"""
Computes mIoU for semantic segmentation
Args:
modalities (list): list of modalities used
num_classes (int): number of classes
save_results (bool): if True saves prediction in a csv file
get_classes (bool): if True returns the classwise F1 Score
"""
def __init__(
self,
modalities: list = [],
num_classes: int = 15,
save_results: bool = False,
get_classes: bool = False,
):
super().__init__()
self.modality = modalities[0]
self.num_classes = num_classes
self.get_classes = get_classes
self.miou = MeanIoU(num_classes=num_classes, per_class=True)
self.save_results = save_results
if save_results:
self.results = {}
def update(self, pred, gt):
label = gt["label"].flatten(0, 1).long()
self.miou(
torch.nn.functional.one_hot(
pred.flatten(2, 3).permute(0, 2, 1).flatten(0, 1).argmax(dim=1),
num_classes=self.num_classes,
),
torch.nn.functional.one_hot(label, num_classes=self.num_classes),
)
if self.save_results:
for i, name in enumerate(gt["name"]):
self.results[name] = list(pred.cpu()[i].numpy())
def compute(self):
if self.get_classes:
miou = self.miou.compute()
out = {"mIoU": sum(miou) / len(miou)}
for i in range(len(miou[:-1])):
out["_".join(["IoU", str(i)])] = miou[i]
return out
miou = self.miou.compute()
out = {
"IoU": miou[1].item(), # IoU of the foreground class
}
if self.save_results:
out["results"] = self.results
return out
return out
class MetricsSemSegJ(Metric):
"""
Computes the IoU for binary segmentation
Args:
modalities (list): list of modalities used
num_classes (int): number of classes
save_results (bool): if True saves prediction in a csv file
get_classes (bool): if True returns the classwise F1 Score
"""
def __init__(
self,
modalities: list = [],
num_classes: int = 15,
save_results: bool = False,
get_classes: bool = False,
save_dir: str = "",
):
super().__init__()
self.modality = modalities[0]
self.num_classes = num_classes
self.get_classes = get_classes
self.miou = JaccardIndex(task="multiclass", num_classes=2, ignore_index=-1)
self.save_results = save_results
if save_results:
self.save_dir = save_dir
if not os.path.exists(save_dir):
os.makedirs(save_dir)
def update(self, pred, gt):
self.miou(
pred.flatten(2, 3).permute(0, 2, 1).flatten(0, 1).argmax(dim=1),
gt["label"].flatten(1, 2).flatten(0, 1).long(),
)
if self.save_results:
for i, name in enumerate(gt["name"]):
np.save(self.save_dir + str(name) + ".npy", pred.cpu()[i].numpy())
np.save(self.save_dir + str(name) + "_gt.npy", gt["label"].cpu()[i].numpy())
def compute(self):
if self.get_classes:
miou = self.miou.compute()
out = {"mIoU": sum(miou[:-1]) / len(miou[:-1])}
for i in range(len(miou[:-1])):
out["_".join(["IoU", str(i)])] = miou[i]
return out
miou = self.miou.compute()
out = {
"IoU": miou,
}
return out
import numpy as np
class Metric(object):
"""Base class for all metrics.
From: https://github.com/pytorch/tnt/blob/master/torchnet/meter/meter.py
"""
def reset(self):
pass
def add(self):
pass
def value(self):
pass
class ConfusionMatrix(Metric):
"""Constructs a confusion matrix for a multi-class classification problems.
Does not support multi-label, multi-class problems.
Keyword arguments:
- num_classes (int): number of classes in the classification problem.
- normalized (boolean, optional): Determines whether or not the confusion
matrix is normalized or not. Default: False.
Modified from: https://github.com/pytorch/tnt/blob/master/torchnet/meter/confusionmeter.py
"""
def __init__(self, num_classes, normalized=False, device="cpu", lazy=True):
super().__init__()
if device == "cpu":
self.conf = np.ndarray((num_classes, num_classes), dtype=np.int64)
else:
self.conf = torch.zeros((num_classes, num_classes)).cuda()
self.normalized = normalized
self.num_classes = num_classes
self.device = device
self.reset()
self.lazy = lazy
def reset(self):
if self.device == "cpu":
self.conf.fill(0)
else:
self.conf = torch.zeros(self.conf.shape).cuda()
def add(self, predicted, target):
"""Computes the confusion matrix
The shape of the confusion matrix is K x K, where K is the number
of classes.
Keyword arguments:
- predicted (Tensor or numpy.ndarray): Can be an N x K tensor/array of
predicted scores obtained from the model for N examples and K classes,
or an N-tensor/array of integer values between 0 and K-1.
- target (Tensor or numpy.ndarray): Can be an N x K tensor/array of
ground-truth classes for N examples and K classes, or an N-tensor/array
of integer values between 0 and K-1.
"""
# If target and/or predicted are tensors, convert them to numpy arrays
if self.device == "cpu":
if torch.is_tensor(predicted):
predicted = predicted.cpu().numpy()
if torch.is_tensor(target):
target = target.cpu().numpy()
assert (
predicted.shape[0] == target.shape[0]
), "number of targets and predicted outputs do not match"
if len(predicted.shape) != 1:
assert (
predicted.shape[1] == self.num_classes
), "number of predictions does not match size of confusion matrix"
predicted = predicted.argmax(1)
else:
if not self.lazy:
assert (predicted.max() < self.num_classes) and (
predicted.min() >= 0
), "predicted values are not between 0 and k-1"
if len(target.shape) != 1:
if not self.lazy:
assert (
target.shape[1] == self.num_classes
), "Onehot target does not match size of confusion matrix"
assert (target >= 0).all() and (
target <= 1
).all(), "in one-hot encoding, target values should be 0 or 1"
assert (target.sum(1) == 1).all(), "multi-label setting is not supported"
target = target.argmax(1)
else:
if not self.lazy:
assert (target.max() < self.num_classes) and (
target.min() >= 0
), "target values are not between 0 and k-1"
# hack for bincounting 2 arrays together
x = predicted + self.num_classes * target
if self.device == "cpu":
bincount_2d = np.bincount(x.astype(np.int64), minlength=self.num_classes**2)
assert bincount_2d.size == self.num_classes**2
conf = bincount_2d.reshape((self.num_classes, self.num_classes))
else:
bincount_2d = torch.bincount(x, minlength=self.num_classes**2)
conf = bincount_2d.view((self.num_classes, self.num_classes))
self.conf += conf
def value(self):
"""
Returns:
Confustion matrix of K rows and K columns, where rows corresponds
to ground-truth targets and columns corresponds to predicted
targets.
"""
if self.normalized:
conf = self.conf.astype(np.float32)
return conf / conf.sum(1).clip(min=1e-12)[:, None]
else:
return self.conf
class IoU(Metric):
"""Computes the intersection over union (IoU) per class and corresponding
mean (mIoU).
Intersection over union (IoU) is a common evaluation metric for semantic
segmentation. The predictions are first accumulated in a confusion matrix
and the IoU is computed from it as follows:
IoU = true_positive / (true_positive + false_positive + false_negative).
Keyword arguments:
- num_classes (int): number of classes in the classification problem
- normalized (boolean, optional): Determines whether or not the confusion
matrix is normalized or not. Default: False.
- ignore_index (int or iterable, optional): Index of the classes to ignore
when computing the IoU. Can be an int, or any iterable of ints.
"""
def __init__(
self,
num_classes,
normalized=False,
ignore_index=None,
cm_device="cpu",
lazy=True,
visu=False,
save_dir="",
):
super().__init__()
self.conf_metric = ConfusionMatrix(num_classes, normalized, device=cm_device, lazy=lazy)
self.lazy = lazy
self.visu = visu
if self.visu:
self.save_dir = save_dir
if not os.path.exists(save_dir):
os.makedirs(save_dir)
if ignore_index is None:
self.ignore_index = None
elif isinstance(ignore_index, int):
self.ignore_index = (ignore_index,)
else:
try:
self.ignore_index = tuple(ignore_index)
except TypeError:
raise ValueError("'ignore_index' must be an int or iterable")
def reset(self):
self.conf_metric.reset()
def update(self, predicted, tg):
"""Adds the predicted and target pair to the IoU metric.
Keyword arguments:
- predicted (Tensor): Can be a (N, K, H, W) tensor of
predicted scores obtained from the model for N examples and K classes,
or (N, H, W) tensor of integer values between 0 and K-1.
- target (Tensor): Can be a (N, K, H, W) tensor of
target scores for N examples and K classes, or (N, H, W) tensor of
integer values between 0 and K-1.
"""
target = tg["label"]
# Dimensions check
assert predicted.size(0) == target.size(
0
), "number of targets and predicted outputs do not match"
assert (
predicted.dim() == 3 or predicted.dim() == 4
), "predictions must be of dimension (N, H, W) or (N, K, H, W)"
assert (
target.dim() == 3 or target.dim() == 4
), "targets must be of dimension (N, H, W) or (N, K, H, W)"
# If the tensor is in categorical format convert it to integer format
if predicted.dim() == 4:
_, predicted = predicted.max(1)
if target.dim() == 4:
_, target = target.max(1)
self.conf_metric.add(predicted.view(-1), target.view(-1))
if self.visu:
for i, name in enumerate(tg["name"]):
np.save(self.save_dir + str(name) + ".npy", predicted.cpu()[i].numpy())
np.save(self.save_dir + str(name) + "_gt.npy", target.cpu()[i].numpy())
def value(self):
"""Computes the IoU and mean IoU.
The mean computation ignores NaN elements of the IoU array.
Returns:
Tuple: (IoU, mIoU). The first output is the per class IoU,
for K classes it's numpy.ndarray with K elements. The second output,
is the mean IoU.
"""
conf_matrix = self.conf_metric.value()
if self.ignore_index is not None:
conf_matrix[:, self.ignore_index] = 0
conf_matrix[self.ignore_index, :] = 0
true_positive = np.diag(conf_matrix)
false_positive = np.sum(conf_matrix, 0) - true_positive
false_negative = np.sum(conf_matrix, 1) - true_positive
# Just in case we get a division by 0, ignore/hide the error
with np.errstate(divide="ignore", invalid="ignore"):
iou = true_positive / (true_positive + false_positive + false_negative)
return iou, np.nanmean(iou)
def compute(self):
conf_matrix = self.conf_metric.value()
if torch.is_tensor(conf_matrix):
conf_matrix = conf_matrix.cpu().numpy()
if self.ignore_index is not None:
conf_matrix[:, self.ignore_index] = 0
conf_matrix[self.ignore_index, :] = 0
true_positive = np.diag(conf_matrix)
false_positive = np.sum(conf_matrix, 0) - true_positive
false_negative = np.sum(conf_matrix, 1) - true_positive
# Just in case we get a division by 0, ignore/hide the error
with np.errstate(divide="ignore", invalid="ignore"):
iou = true_positive / (true_positive + false_positive + false_negative)
miou = float(np.nanmean(iou))
acc = float(np.diag(conf_matrix).sum() / conf_matrix.sum() * 100)
out = {"mIoU": miou, "acc": acc}
return out
class MetricsBinarySemSeg(Metric):
"""
Computes IoU Score for binary segmentation tasks
Args:
modalities (list): list of modalities used
save_results (bool): if True saves prediction in a csv file
threshold (float): threshold for binary prediction (default: 0.5)
"""
def __init__(self, modalities: list = [], save_results: bool = False, threshold: float = 0.5):
super().__init__()
self.modality = modalities[0]
self.threshold = threshold
self.miou = MeanIoU(num_classes=2, per_class=True) # Binary: 2 classes (0 and 1)
self.save_results = save_results
if save_results:
self.results = {}
def update(self, pred, gt):
# Convert predictions to binary using threshold
pred_binary = (pred.sigmoid() > self.threshold).float()
# Convert to one-hot encoding
pred_one_hot = torch.nn.functional.one_hot(
pred_binary.flatten(2, 3).permute(0, 2, 1).flatten(0, 1).long(), num_classes=2
)
gt_one_hot = torch.nn.functional.one_hot(
gt["label"].flatten(1, 2).flatten(0, 1).long(), num_classes=2
)
self.miou(pred_one_hot, gt_one_hot)
if self.save_results:
for i, name in enumerate(gt["name"]):
self.results[name] = pred_binary.cpu()[i].numpy()
def compute(self):
miou = self.miou.compute()
# For binary segmentation, we typically care about IoU of class 1 (foreground)
out = {
"mIoU": miou[1].item(), # IoU of the foreground class
"IoU_background": miou[0].item(),
"IoU_foreground": miou[1].item(),
}
if self.save_results:
out["results"] = self.results
return out
class MetricsReg(Metric):
"""
Computes the Root Mean Square Error (RMSE) for regression tasks by applying a softplus activation
to the predictions before computing the MSE loss.
Args:
modalities (list): List of modalities used in the model. Currently not used in the implementation
but kept for consistency with other metric classes.
Attributes:
softplus (nn.Softplus): Softplus activation function applied to predictions
mse (torch.Tensor): Accumulated mean squared error
total_samples (torch.Tensor): Counter for total number of samples processed
"""
def __init__(
self,
modalities: list = [],
):
super().__init__()
self.softplus = nn.Softplus()
self.mse = torch.tensor(0.0)
self.total_samples = torch.tensor(0)
def update(self, pred, gt):
self.mse += F.mse_loss(self.softplus(pred), gt["label"]).cpu()
self.total_samples += 1
def compute(self):
rmse = torch.sqrt(self.mse / self.total_samples)
out = {"RMSE": rmse.item()}
return out
class SegPangaea(Metric):
"""
SegPangaea is a class for evaluating segmentation models using a confusion matrix approach.
Attributes:
num_classes (int): Number of classes in the segmentation task
ignore_index (int): Index value to ignore when computing metrics
confusion_matrix (torch.Tensor): Matrix of shape (num_classes, num_classes) to store predictions
Methods:
update(pred, gt):
Updates the confusion matrix with new predictions and ground truth.
Args:
pred (torch.Tensor): Model predictions
gt (dict): Dictionary containing ground truth labels under 'label' key
compute():
Computes various metrics from the accumulated confusion matrix.
Returns:
dict: Dictionary containing the following metrics:
- mIoU: Mean Intersection over Union across all classes
- mF1: Mean F1 score across all classes
- mAcc: Mean pixel accuracy
"""
def __init__(self, num_classes, ignore_index):
self.num_classes = num_classes
self.ignore_index = ignore_index
self.confusion_matrix = torch.zeros(num_classes, num_classes)
def update(self, pred, gt):
label = gt["label"].flatten(1, 2)
pred = torch.argmax(pred, dim=1).flatten(1, 2)
valid_mask = label != self.ignore_index
pred, target = pred[valid_mask], label[valid_mask]
count = torch.bincount((pred * self.num_classes + target), minlength=self.num_classes**2)
self.confusion_matrix = self.confusion_matrix.to(pred.device)
self.confusion_matrix += count.view(self.num_classes, self.num_classes)
def compute(self):
# Calculate IoU for each class
intersection = torch.diag(self.confusion_matrix)
union = self.confusion_matrix.sum(dim=1) + self.confusion_matrix.sum(dim=0) - intersection
iou = intersection / (union + 1e-6)
# Calculate precision and recall for each class
precision = intersection / (self.confusion_matrix.sum(dim=0) + 1e-6)
recall = intersection / (self.confusion_matrix.sum(dim=1) + 1e-6)
# Calculate F1-score for each class
f1 = 2 * (precision * recall) / (precision + recall + 1e-6)
# Calculate mean IoU, mean F1-score, and mean Accuracy
miou = iou.mean().item()
mf1 = f1.mean().item()
macc = (intersection.sum() / (self.confusion_matrix.sum() + 1e-6)).item()
# Convert metrics to CPU and to Python scalars
iou = iou.cpu()
f1 = f1.cpu()
precision = precision.cpu()
recall = recall.cpu()
# Prepare the metrics dictionary
metrics = {
"mIoU": miou,
"mF1": mf1,
"mAcc": macc,
}
return metrics
|