kaiko-eva 0.1.1__py3-none-any.whl → 0.1.3__py3-none-any.whl

eva/vision/models/networks/decoders/segmentation/semantic/with_image.py

@@ -0,0 +1,94 @@
+"""Convolutional semantic segmentation decoders that use input image & feature maps as input."""
+
+from typing import List
+
+import torch
+from torch import nn
+from torchvision.transforms.functional import rgb_to_grayscale
+from typing_extensions import override
+
+from eva.vision.models.networks.decoders.segmentation import decoder2d
+from eva.vision.models.networks.decoders.segmentation.typings import DecoderInputs
+
+
+class ConvDecoderWithImage(decoder2d.Decoder2D):
+    """A convolutional that in addition to encoded features, also takes the input image as input.
+
+    In a first stage, the input features are upsampled and passed through a convolutional layer,
+    while in the second stage, the input image channels are concatenated with the upsampled features
+    and passed through additional convolutional blocks in order to combine the image prior
+    information with the encoded features. Lastly, a 1x1 conv operation reduces the number of
+    channels to the number of classes.
+    """
+
+    _default_hidden_dims = [64, 32, 32]
+
+    def __init__(
+        self,
+        in_features: int,
+        num_classes: int,
+        greyscale: bool = False,
+        hidden_dims: List[int] | None = None,
+    ) -> None:
+        """Initializes the decoder.
+
+        Args:
+            in_features: The hidden dimension size of the embeddings.
+            num_classes: Number of output classes as channels.
+            greyscale: Whether to convert input images to greyscale.
+            hidden_dims: List of hidden dimensions for the convolutional layers.
+        """
+        hidden_dims = hidden_dims or self._default_hidden_dims
+        if len(hidden_dims) != 3:
+            raise ValueError("Hidden dims must have 3 elements.")
+
+        super().__init__(
+            layers=nn.Sequential(
+                nn.Upsample(scale_factor=2),
+                Conv2dBnReLU(in_features, hidden_dims[0]),
+            )
+        )
+        self.greyscale = greyscale
+
+        additional_hidden_dims = 1 if greyscale else 3
+        self.image_block = nn.Sequential(
+            Conv2dBnReLU(hidden_dims[0] + additional_hidden_dims, hidden_dims[1]),
+            Conv2dBnReLU(hidden_dims[1], hidden_dims[2]),
+        )
+
+        self.classifier = nn.Conv2d(hidden_dims[2], num_classes, kernel_size=1)
+
+    @override
+    def forward(self, decoder_inputs: DecoderInputs) -> torch.Tensor:
+        if decoder_inputs.images is None:
+            raise ValueError("Input images are missing.")
+
+        logits = super().forward(decoder_inputs)
+        in_images = (
+            rgb_to_grayscale(decoder_inputs.images) if self.greyscale else decoder_inputs.images
+        )
+        logits = torch.cat([logits, in_images], dim=1)
+        logits = self.image_block(logits)
+
+        return self.classifier(logits)
+
+
+class Conv2dBnReLU(nn.Sequential):
+    """A single convolutional layer with batch normalization and ReLU activation."""
+
+    def __init__(
+        self, in_channels: int, out_channels: int, kernel_size: int = 3, padding: int = 1
+    ) -> None:
+        """Initializes the layer.
+
+        Args:
+            in_channels: Number of input channels.
+            out_channels: Number of output channels.
+            kernel_size: Size of the convolutional kernel.
+            padding: Padding size for the convolutional layer.
+        """
+        super().__init__(
+            nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=padding),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(inplace=True),
+        )

eva/vision/models/networks/decoders/segmentation/typings.py

@@ -0,0 +1,18 @@
+"""Type-hints for segmentation decoders."""
+
+from typing import List, NamedTuple, Tuple
+
+import torch
+
+
+class DecoderInputs(NamedTuple):
+    """Input scheme for segmentation decoders."""
+
+    features: List[torch.Tensor]
+    """List of image features generated by the encoder from the original images."""
+
+    image_size: Tuple[int, int]
+    """Size of the original input images to be used for upsampling."""
+
+    images: torch.Tensor | None = None
+    """The original input images for which the encoder generated the encoded_images."""

eva/vision/utils/io/__init__.py

@@ -2,7 +2,12 @@
 
 from eva.vision.utils.io.image import read_image, read_image_as_array, read_image_as_tensor
 from eva.vision.utils.io.mat import read_mat, save_mat
-from eva.vision.utils.io.nifti import fetch_nifti_shape, read_nifti, save_array_as_nifti
+from eva.vision.utils.io.nifti import (
+    fetch_nifti_axis_direction_code,
+    fetch_nifti_shape,
+    read_nifti,
+    save_array_as_nifti,
+)
 from eva.vision.utils.io.text import read_csv
 
 __all__ = [
@@ -10,6 +15,7 @@ __all__ = [
     "read_image_as_array",
     "read_image_as_tensor",
     "fetch_nifti_shape",
+    "fetch_nifti_axis_direction_code",
     "read_nifti",
     "save_array_as_nifti",
     "read_csv",

eva/vision/utils/io/nifti.py

@@ -5,6 +5,7 @@ from typing import Any, Tuple
 import nibabel as nib
 import numpy as np
 import numpy.typing as npt
+from nibabel import orientations
 
 from eva.vision.utils.io import _utils
 
@@ -28,13 +29,13 @@ def read_nifti(
         ValueError: If the input channel is invalid for the image.
     """
     _utils.check_file(path)
-    image_data = nib.load(path)  # type: ignore
+    image_data: nib.Nifti1Image = nib.load(path)  # type: ignore
     if slice_index is not None:
-        image_data = image_data.slicer[:, :, slice_index : slice_index + 1]  # type: ignore
+        image_data = image_data.slicer[:, :, slice_index : slice_index + 1]
 
-    image_array = image_data.get_fdata()  # type: ignore
+    image_array = image_data.get_fdata()
     if use_storage_dtype:
-        image_array = image_array.astype(image_data.get_data_dtype())  # type: ignore
+        image_array = image_array.astype(image_data.get_data_dtype())
 
     return image_array
 
@@ -73,3 +74,17 @@ def fetch_nifti_shape(path: str) -> Tuple[int]:
     _utils.check_file(path)
     image = nib.load(path)  # type: ignore
     return image.header.get_data_shape()  # type: ignore
+
+
+def fetch_nifti_axis_direction_code(path: str) -> str:
+    """Fetches the NIfTI axis direction code from a file.
+
+    Args:
+        path: The path to the NIfTI file.
+
+    Returns:
+        The axis direction codes as string (e.g. "LAS").
+    """
+    _utils.check_file(path)
+    image_data: nib.Nifti1Image = nib.load(path)  # type: ignore
+    return "".join(orientations.aff2axcodes(image_data.affine))

{kaiko_eva-0.1.1.dist-info → kaiko_eva-0.1.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: kaiko-eva
-Version: 0.1.1
+Version: 0.1.3
 Summary: Evaluation Framework for oncology foundation models.
 Keywords: machine-learning,evaluation-framework,oncology,foundation-models
 Author-Email: Ioannis Gatopoulos <ioannis@kaiko.ai>, =?utf-8?q?Nicolas_K=C3=A4nzig?= <nicolas@kaiko.ai>, Roman Moser <roman@kaiko.ai>
@@ -468,41 +468,10 @@ and [tutorials](https://kaiko-ai.github.io/eva/dev/user-guide/advanced/replicate
 
 ## Leaderboards
 
-In this section you will find model benchmarks which were generated with _`eva`_.
+The following table shows the FMs we have evaluated with _`eva`_. For more detailed information about the evaluation process, please refer to our [documentation](https://kaiko-ai.github.io/eva/main/leaderboards/).
 
-### Table I: WSI and microscopy image tasks
+![Pathology Leaderboard](./docs/images/leaderboard.svg)
 
-<br />
-
-<div align="center">
-
-| Model   | BACH  | CRC   | MHIST | PCam   | Camelyon16 | PANDA | CoNSeP | MoNuSAC |
-|---------|-------|-------|-------|--------|------------|-------|------------|-------|
-| ViT-S/16 _(random)_	<sup>[1]</sup> | 0.411|0.613|0.5|0.752|0.551|0.347|0.489|0.394|
-| ViT-S/16 _(ImageNet)_ <sup>[1]</sup> | 0.675|0.936|0.827|0.861|0.751|0.676|0.54|0.512|
-| DINO<sub>(p=16)</sub> <sup>[2]</sup> | 0.77|0.936|0.751|0.905|0.869|0.737|0.625|0.549|
-| Phikon <sup>[3]</sup>                | 0.715|0.942|0.766|0.925|0.879|0.784|0.68|0.554|
-| UNI <sup>[4]</sup>                   | 0.797|0.95|0.835|0.939|0.933|0.774|0.67|0.575|
-| ViT-S/16 _(kaiko.ai)_ <sup>[5]</sup> | 0.8|0.949|0.831|0.902|0.897|0.77|0.622|0.573|
-| ViT-S/8 _(kaiko.ai)_ <sup>[5]</sup>  | 0.825|0.948|0.826|0.887|0.879|0.741|0.677|0.617|
-| ViT-B/16 _(kaiko.ai)_	<sup>[5]</sup> | 0.846|0.959|0.839|0.906|0.891|0.753|0.647|0.572|
-| ViT-B/8 _(kaiko.ai)_ <sup>[5]</sup>  | 0.867|0.952|0.814|0.921|0.939|0.761|0.706|0.661|
-| ViT-L/14 _(kaiko.ai)_ <sup>[5]</sup> | 0.862|0.935|0.822|0.907|0.941|0.769|0.686|0.599|
-
-_Table I: Linear probing evaluation of FMs on patch-level downstream datasets.<br> We report balanced accuracy
- for classification tasks and generalized Dice score for semgetnation tasks, averaged over 5 runs. Results are 
- reported on the "test" split if available and otherwise on the "validation" split._
-
-</div>
-
-<br />
-
-_References_:
-1. _"Emerging properties in self-supervised vision transformers”_, [arXiv](https://arxiv.org/abs/2104.14294)
-2. _"Benchmarking self-supervised learning on diverse pathology datasets”_, [arXiv](https://arxiv.org/abs/2212.04690)
-3. _"Scaling self-supervised learning for histopathology with masked image modeling”_, [medRxiv](https://www.medrxiv.org/content/10.1101/2023.07.21.23292757v1)
-4. _"A General-Purpose Self-Supervised Model for Computational Pathology”_, [arXiv](https://arxiv.org/abs/2308.15474)
-5. _"Towards Training Large-Scale Pathology Foundation Models: from TCGA to Hospital Scale”_, [arXiv](https://arxiv.org/pdf/2404.15217)
 
 ## Contributing
 

{kaiko_eva-0.1.1.dist-info → kaiko_eva-0.1.3.dist-info}/RECORD

@@ -7,7 +7,7 @@ eva/core/callbacks/config.py,sha256=-DRt20a2aF9Z9-7nZvbGBcOZ30qNf3ESf25EPRgRL1w,
 eva/core/callbacks/writers/__init__.py,sha256=z8cNVJOanj-yYyIiX-mOkhur1NExuCOKzPjp4mmm3AE,232
 eva/core/callbacks/writers/embeddings/__init__.py,sha256=zMxP4POf1yFFUVSjGcsQgyPYNY6JsZO_F66ngKJZaG8,301
 eva/core/callbacks/writers/embeddings/_manifest.py,sha256=pB9nGv9ofVbnI4HHPrHY4p7jeFiFQTRc09zszto_DTk,2424
-eva/core/callbacks/writers/embeddings/base.py,sha256=rdoCtMuegiO6Gb0vM8a-KGI33Eq0hS0Qnlh-1Y6_96s,7593
+eva/core/callbacks/writers/embeddings/base.py,sha256=YrjdqIFBYla-5jleUUTyJ5OHmvInhzLLHXEGmzwa9xA,7620
 eva/core/callbacks/writers/embeddings/classification.py,sha256=pYU0dD08IXH4_pK-P43LqCMD17X-AP5Ayo9gbovwv6U,4522
 eva/core/callbacks/writers/embeddings/segmentation.py,sha256=6AMGfD9Ur35TyH8ztjPx4ayB1Kbywymmu9LriMGLDoY,3135
 eva/core/callbacks/writers/embeddings/typings.py,sha256=qXZSlasaSKHad6HyJeRTeUv-ZeJVM-R3oIaasD8ZPc8,997
@@ -17,7 +17,7 @@ eva/core/cli/logo.py,sha256=x6-vGWI0s9gza-xxQrBDi2wneb2wFU_mQGHgpAiq2MQ,786
 eva/core/cli/setup.py,sha256=kR-7l4X5Hu8kSLoQZGYGIeLXtn9S_EU52dauDy6fm0w,2663
 eva/core/data/__init__.py,sha256=yG3BeOWhp1EjVYMFqx8M_TBWFDyfIwwksQGQmMdSPaI,340
 eva/core/data/dataloaders/__init__.py,sha256=fbNClVZ8J3QoGi4qiPq635ig1j9GdI7six3RhfwDbjY,110
-eva/core/data/dataloaders/dataloader.py,sha256=-mWFFLtem1Ijbi8XGveFSv5XzUU7SyKwiT5Ahikzghw,2368
+eva/core/data/dataloaders/dataloader.py,sha256=n2NNlWKurEC1TVis347l6YmPgNOGTaoxwIyoClyPt7s,2383
 eva/core/data/datamodules/__init__.py,sha256=qZchYbgxo9lxYnGoqdk0C6MfS2IbF0WItO0kCdP9Mqc,229
 eva/core/data/datamodules/call.py,sha256=jjj9w3UXYuQB-qyCcw1EZpRJW10OC1I3dvgvsuQWLck,940
 eva/core/data/datamodules/datamodule.py,sha256=dclC2YJAXUGEUpV9ZRWQS43-ksFIPgVeFudsyrj9kdc,3878
@@ -32,8 +32,8 @@ eva/core/data/datasets/embeddings.py,sha256=zNEO8KxqiOopcN_lTjwtEAm9xbnYDSjOE8X2
 eva/core/data/samplers/__init__.py,sha256=WikBo1DemCx6o2vFfNwSwODlmCT2zWUXtCNwiWCVAFE,100
 eva/core/data/samplers/sampler.py,sha256=vrrXERWC67fjmTk_uwD7s9-8-rdhvnx7OlSipHE6sdY,119
 eva/core/data/splitting/__init__.py,sha256=VQJ8lfakbv6y2kAk3VDtITAvh7kcZo3H1JwJBc5jT08,198
-eva/core/data/splitting/random.py,sha256=gmweyGtL4rbWoUaH1q63LjKpT-TCwk2mdB4Vw_jLTQo,1353
-eva/core/data/splitting/stratified.py,sha256=_1Eh6QuupxLexrABpwNXiDxDisoTiK8axsV3nvQXCx0,1915
+eva/core/data/splitting/random.py,sha256=r6iy7j34seRTlyB79_Xy7m6lsKRi8ZM9X5Ln1b-SBjg,1453
+eva/core/data/splitting/stratified.py,sha256=dliRHgEyZTOzpJrR8FzaRaAnjUxu_VsZUuy_4MPQjlY,2265
 eva/core/data/transforms/__init__.py,sha256=n0TczmJSc9EjR6JezAZqlZIN4Gz_X3UBePbyDSC7JkE,308
 eva/core/data/transforms/dtype/__init__.py,sha256=r_LM_hdh_gTsrgh3shDTdMpu-lgQNHJ1yD6wY3omPyg,174
 eva/core/data/transforms/dtype/array.py,sha256=RDSkXlnSHSYyU_gv7vw33OZ7vhEy62PQGoE3htGGaqc,725
@@ -51,17 +51,15 @@ eva/core/loggers/log/image.py,sha256=iUwntQCdRNLtkSdqu8CvV34l06zPYVo4NAW2gUeiJIM
 eva/core/loggers/log/parameters.py,sha256=7Xi-I5gQvEVv71d58bwdZ-Hb4287NXxaUyMfriq_KDU,1634
 eva/core/loggers/log/utils.py,sha256=k4Q7uKpAQctfDv0EEYPnPv6wt9LnckEeqGvbYSLfKO0,415
 eva/core/loggers/loggers.py,sha256=igHxdxJSotWSg6nEOKnfFuBszzblHgi8T7sBrE00FEs,166
-eva/core/metrics/__init__.py,sha256=CtmUcB2bh-JlI0wOvSwwPFB1OzaqsSM3dPHVQh7hMXY,714
+eva/core/losses/__init__.py,sha256=D-Mp9fUFFFoH9YYWntVH3B839zHS3GjFJzkbQThzj6Y,118
+eva/core/losses/cross_entropy.py,sha256=Sunz7ogDAJpGvZtuk9cAxKZJBO08CKIEvbCoewEvees,862
+eva/core/metrics/__init__.py,sha256=-9Qch4npEQpy3oF6NUhh9WinCmFBFe0D2eEYCR0S0xU,558
 eva/core/metrics/average_loss.py,sha256=AyFOnCXBD5T62eSYf6eGAAJsqt8x-KaHgc8OLkCHjzE,1267
 eva/core/metrics/binary_balanced_accuracy.py,sha256=MabsXAtVfLqSaSIIpE0HIM6bo8uRszl6obueHI6vJi0,806
-eva/core/metrics/defaults/__init__.py,sha256=uPQzkna6Lb0VnCtC4IEPSB9d5jI1_0SohjUOMSo3o1Q,373
+eva/core/metrics/defaults/__init__.py,sha256=IeqLcoxLNddtuuFao5r85ZVxTyFP6iFsj1K8iXCTSes,255
 eva/core/metrics/defaults/classification/__init__.py,sha256=xMzE4yV8NoUdcmk2FCKohEUav1GJcxYn60S1KNgXbJY,316
 eva/core/metrics/defaults/classification/binary.py,sha256=9ll6ZOcNGQdsg7ln9DAQ0u-OzsXSzEbueXe-dVJkJZ8,2322
 eva/core/metrics/defaults/classification/multiclass.py,sha256=8Aesy_rKtp4KxfXJtDCmk6FsGxIFS4Ywu2CH1VIRL7M,2518
-eva/core/metrics/defaults/segmentation/__init__.py,sha256=n6gDc603uRWOByAAPFkmZiPH2rEoZ3lSV9MC4nRMBuc,189
-eva/core/metrics/defaults/segmentation/multiclass.py,sha256=_M7NtvwIrfzLLXtAYflFjIle6UeHYU9TwWo3IHl0wlw,1715
-eva/core/metrics/generalized_dice.py,sha256=28vTdmh6QyLfSGtT5oARXp2Hd58EBNg5G0dSBfctvcY,2271
-eva/core/metrics/mean_iou.py,sha256=eAvAe1BiYEXjOtWHUZD_5hBGuRmNhHVYuyGls8YC-1g,4619
 eva/core/metrics/structs/__init__.py,sha256=cvn7E4k5vJmpwJj_zezmtZa_Nl_RddDM1G-MO8TP0po,422
 eva/core/metrics/structs/collection.py,sha256=bNfCekHN8pzD49-YTqVxrmxFtiQfNxnv-RwkxCL6rbc,149
 eva/core/metrics/structs/metric.py,sha256=zdnE0ZVTSYAMl7rW_OL6e1XiZDvLTirYqV0lgJCleXY,109
@@ -72,7 +70,7 @@ eva/core/models/__init__.py,sha256=bQSpfQJKuDMWosjcMhP7t5jdOSV6OyxdxTOIW9w1woE,3
 eva/core/models/modules/__init__.py,sha256=QJWJ42BceXZBzDGgk5FHBcCaRrB9egTFKVF6gDsBYfM,255
 eva/core/models/modules/head.py,sha256=iHrEOjYfshFI6OdXxJJTZyfCoUs2fimitINNcB6ENsc,4321
 eva/core/models/modules/inference.py,sha256=ih-0Rr2oNf2N6maiXPOW7XH5KVwUT1_MOxnJKOhJ1uQ,978
-eva/core/models/modules/module.py,sha256=7mCzyvBNOWhvN8sNa91yB79iSBlJlYh9sypL37Nwdes,6836
+eva/core/models/modules/module.py,sha256=LtjYxTZb7UY0owonmt_yQ5EySw3sX-xD9HLN2io8EK4,6697
 eva/core/models/modules/typings.py,sha256=yFMJCE4Nrfd8VEXU1zk8p6Sz5M7UslwitYPVC2OPLSY,776
 eva/core/models/modules/utils/__init__.py,sha256=pnbxlEhT87JimWNr-NSNCv7VNR-IyDi_A9qRWmvlzwQ,227
 eva/core/models/modules/utils/batch_postprocess.py,sha256=RwnDcjJy3uvVirpgx_80Q2CUYKfJKipVwjyX7AF2CKw,3088
@@ -80,8 +78,8 @@ eva/core/models/modules/utils/grad.py,sha256=bl8qb8g4Nhg1KAGfbEV_9HTKkoT0azRwfs9
 eva/core/models/networks/__init__.py,sha256=yqx6UmG1Eg3vb1O_tnK_axnJWabEl9ULkDWiPN440Xc,85
 eva/core/models/networks/mlp.py,sha256=thk-x4pviE3fCaMW9k3I2Oe5_DxfC-CqUrtolvVdXug,2418
 eva/core/models/transforms/__init__.py,sha256=oYL3gNUUKZFViTu6GT1jVE2Kv1xFYPuyiYp-sErtVVg,257
-eva/core/models/transforms/extract_cls_features.py,sha256=odtqawFoDZZCvCg0bp8G8PlUY8KrPAQBZsNOcTZv02E,1081
-eva/core/models/transforms/extract_patch_features.py,sha256=41zCkX-ls-rvqB4B4kE5_lWMNhec65yatdDNa0yjRf0,1751
+eva/core/models/transforms/extract_cls_features.py,sha256=tFRd4H-eGFIGCfZt6wuZGibDmAoNXKSsn15bBw0IDdc,1482
+eva/core/models/transforms/extract_patch_features.py,sha256=k50jTLPWxbfvciH9QZSzTAGqWwDSVpXAteme_Qg2d6E,2202
 eva/core/models/wrappers/__init__.py,sha256=P-ipr4NtKqPU6ubAjKLGxFf1Qt2yDSNtgS2Xz5sBahQ,364
 eva/core/models/wrappers/_utils.py,sha256=HXUyGcILaa8GK31ViIHCKRU4f9kbjAPYQmhvN2N7jSc,957
 eva/core/models/wrappers/base.py,sha256=xKMUSXk93wI67p_wmh7jujK-bxvIefO1noYaAJN_5Ak,1359
@@ -102,6 +100,7 @@ eva/core/utils/memory.py,sha256=ZvcbS1eUPXdHIoL8ctFU56_-cyUniObBmIctUbvso48,636
 eva/core/utils/multiprocessing.py,sha256=PxUxMyvI62lghyWF46O5RNL-J7DUR2IrXSwdkbhC0ic,1383
 eva/core/utils/operations.py,sha256=eoC_ScuHUMDCuk08j1bosiQZdPrgiIODqqheR9MtJHQ,641
 eva/core/utils/parser.py,sha256=2czmwEGJJ6PtmaD86s9I14P-_sek4DmDCkEatRGT5sI,725
+eva/core/utils/progress_bar.py,sha256=KvvsM_v3_Fhb4JvbEEPHb4PJMokg6mNLj-o6dkfzcMc,499
 eva/core/utils/workers.py,sha256=hfx63M82qNg0Dwhre2tl53MnhtRsV7APaDONM9nhVB8,634
 eva/vision/__init__.py,sha256=oUZXFYjwtkWzi8An0uS5Xc84pLKintlXe2iti8zW6BQ,480
 eva/vision/callbacks/__init__.py,sha256=su1V73L0dDVYWSyvV_lnWbszDi2KikRraF7OsgeaKl4,139
@@ -110,30 +109,31 @@ eva/vision/callbacks/loggers/batch/__init__.py,sha256=DVYP7Aonbi4wg_ERHRj_8kb87E
 eva/vision/callbacks/loggers/batch/base.py,sha256=hcAd5iiHvjZ0DIf4Qt4ENT54D6ky_1OO4rKQZqeo-1k,3628
 eva/vision/callbacks/loggers/batch/segmentation.py,sha256=PbgBVp6TGgko7Um8gN0fHyCs2sE42Uqe3M4grxSBykE,6749
 eva/vision/data/__init__.py,sha256=aoKPmX8P2Q2k2W3nlq8vFU41FV6Sze-0SDuWtU-ETh4,111
-eva/vision/data/datasets/__init__.py,sha256=t0pZhs3z-QFHERY5N8FVMQex8TDVG5kfcpGODdUxk8Y,836
+eva/vision/data/datasets/__init__.py,sha256=COhMRB9QJcjfbmfpRcYEztDwN9pl7IJNiH29pCZo4CA,908
 eva/vision/data/datasets/_utils.py,sha256=epPcaYE4w2_LtUKLLQJh6qQxUNVBe22JA06k4WUerYQ,1430
 eva/vision/data/datasets/_validators.py,sha256=77WZj8ewsuxUjW5WegJ-7zDuR6WdF5JbaOYdywhKIK4,2594
-eva/vision/data/datasets/classification/__init__.py,sha256=ht5UPPgP736dt_L1Hb5rJtQnzKJHIhpBnqm3b4BMCZE,663
+eva/vision/data/datasets/classification/__init__.py,sha256=T2eg8k3xxd_Pdbrr7TGYICSo7BVOTMOs1bL-rLnMmro,693
 eva/vision/data/datasets/classification/bach.py,sha256=kZba1dQlJWZAmA03akJ4fVUU-y9W8ezOwlgs2zL-QrE,5432
 eva/vision/data/datasets/classification/base.py,sha256=Ci0HoOhOuHwICTi1TUGA1PwZe642RywolTVfMhKrFHk,2772
-eva/vision/data/datasets/classification/camelyon16.py,sha256=sToajukdw-_V_YO6lbcZToMSLKEjeKxJfjZ8iSdzn-M,8136
+eva/vision/data/datasets/classification/camelyon16.py,sha256=sChvRo0jbOVUMJvfpsFxgFOsYgci3v9wjeMBEjUysJU,8287
 eva/vision/data/datasets/classification/crc.py,sha256=8qjz9OklLg1gAr46RKZdlClmlO9awwfp0dkTs8v5jTE,5670
 eva/vision/data/datasets/classification/mhist.py,sha256=xzShPncSfAV6Q5ojfimeq748MfA0n77fGWa9EpdRzYU,3055
-eva/vision/data/datasets/classification/panda.py,sha256=6VpCsotdksAZSfdD9zcM96Ihr6FshnIgZPZkkt0oSLI,6853
+eva/vision/data/datasets/classification/panda.py,sha256=BU_gDoX3ZSDUugwaO2n0XSZhzseK1rkPoHMRoJLGL84,7303
 eva/vision/data/datasets/classification/patch_camelyon.py,sha256=fElKteZKx4M6AjylnhhgNH1jewHegWc1K8h4FFKp0gE,7171
-eva/vision/data/datasets/classification/wsi.py,sha256=Y8yaPM5qVi13YyRKIcYrRaxmV_yRW8Dl9rj_1kRJ33I,3948
-eva/vision/data/datasets/segmentation/__init__.py,sha256=_E1K8Ld829jVlZ0VcjUy0HP-8aHu4v9rEbTFt3R8O9M,694
+eva/vision/data/datasets/classification/wsi.py,sha256=x3mQ8iwyiSdfQOjJuV7_cd8-LRjjhY9tjtzuD8O87Lg,4099
+eva/vision/data/datasets/segmentation/__init__.py,sha256=hGNr7BM_StxvmlOKWWfHp615qgsrB6BB3qMOiYhE0Og,791
 eva/vision/data/datasets/segmentation/_utils.py,sha256=ps1qpuEkPgvwUw6H-KKaLaYqDBGmN7dNGk3bnS1l6sI,1261
 eva/vision/data/datasets/segmentation/base.py,sha256=11IMODMB7KJ8Bs5p7MyOsBXCyPFJXfYcDLAIMitUwEk,3023
 eva/vision/data/datasets/segmentation/bcss.py,sha256=NHjHd1tgIfIw6TxsZTGb63iMEwXFbWX_JAwRT5WVsj4,8274
-eva/vision/data/datasets/segmentation/consep.py,sha256=mUUGqS1HkUkL1u45LY0rEjcAK0Dawc8abUmFgYEZ_ag,5871
+eva/vision/data/datasets/segmentation/consep.py,sha256=dCD8VsZSvI3-RbHHHAwGWfsNwOJCPkzHpVtzrcuAEVo,5871
 eva/vision/data/datasets/segmentation/embeddings.py,sha256=0KaadzPxN6OrKNnFu3YsGBFkG6XqqvkOZYUhERPwL4A,1220
-eva/vision/data/datasets/segmentation/lits.py,sha256=_9qdjKnYe5YsJ6_UAIrPwMeqoKHyHYmB7q-6uvXqdLQ,6246
-eva/vision/data/datasets/segmentation/monusac.py,sha256=vbXo-T3Rdu_zGja81ZbOimjZMlx2CnRZsC5nH-Dqkyg,8368
-eva/vision/data/datasets/segmentation/total_segmentator_2d.py,sha256=h2daCbFZPm48GjuOAOy0-Cd-WKFkFvus1ZWuoJZY9D4,13070
+eva/vision/data/datasets/segmentation/lits.py,sha256=_R5AGFX8jVPwK3UKaYQfIRLBpM5ZmDg6KRziisUDYps,7175
+eva/vision/data/datasets/segmentation/lits_balanced.py,sha256=s5kPfqB41Vkcm5Jh34mLAO0NweMSIlV2fMXJsRjJsF8,3384
+eva/vision/data/datasets/segmentation/monusac.py,sha256=OTWHAD1b48WeT6phVf466w_nJUOGdBCGKWiWw68PAdw,8423
+eva/vision/data/datasets/segmentation/total_segmentator_2d.py,sha256=pqrNRQu5kbKSd-l5jwaiE67qyF2jLQ3JrO7TjhGGF7w,13098
 eva/vision/data/datasets/structs.py,sha256=RaTDW-B36PumcR5gymhCiX-r8GiKqIFcjqoEEjjFyUE,389
 eva/vision/data/datasets/vision.py,sha256=hKKFMb65UJQzOyYm8FTGkOGBOinMRu7R8sOFMbCmQX4,1100
-eva/vision/data/datasets/wsi.py,sha256=JauEeQEC3niyivLa4FcI4X5GKvDRVpwY6BknzN-vKAQ,6611
+eva/vision/data/datasets/wsi.py,sha256=-rypkcd6CPBM_oPuLszUx9q4zSPzeO1H6JKqvOtLlHw,8282
 eva/vision/data/transforms/__init__.py,sha256=WeFii6JwB0CiOOGLR3tkgAoKgRdmOf2lm0Dadixn8OI,260
 eva/vision/data/transforms/common/__init__.py,sha256=6tvxUgb8wfhgvqejMVulwqssHTJLF7f4_vpf44kxgxY,234
 eva/vision/data/transforms/common/resize_and_clamp.py,sha256=f9-YIX0S9GMAXHP7TWlyRlGfZIVvHgoBHqQ8PzaKbKs,1736
@@ -150,56 +150,69 @@ eva/vision/data/wsi/backends/openslide.py,sha256=VPVJDb6iAe0ZIdYbyFfPLDzHvku8PZX
 eva/vision/data/wsi/backends/pil.py,sha256=CqCWP1ViwpQyVKGLUoEtc4tCHXSAdQpMn6ZX2lNBMns,1403
 eva/vision/data/wsi/backends/tiffslide.py,sha256=f1xOiD4kpL0oRe3xFNT7BM2zYTWBduqL99skk-ZFRwE,1217
 eva/vision/data/wsi/patching/__init__.py,sha256=vSGyui2TkaJpw_wQJldP0Llnym5X9XgK17nuz7S5Hh8,189
-eva/vision/data/wsi/patching/coordinates.py,sha256=IzuF4i63bJYqdJH7eWQYR2q5QHw-80iV6QLibac6CWg,3475
+eva/vision/data/wsi/patching/coordinates.py,sha256=TnsRafUJzsjvfhuP0vvDA294G0f4hu5LqICYNnMFUMo,3870
 eva/vision/data/wsi/patching/mask.py,sha256=o_S4YRdbfaxKCG1_T2skswDirmlzHzVC5exaDJucvD0,4986
 eva/vision/data/wsi/patching/samplers/__init__.py,sha256=QkBbjnZf7IcEPm-ON9SeZP0I3DXUA3pY87dKXXdelz4,458
-eva/vision/data/wsi/patching/samplers/_utils.py,sha256=aJI3mSJjfsMm4eNCAqIwMuXX0mGHl0WUa1vbC0DbbmY,1431
+eva/vision/data/wsi/patching/samplers/_utils.py,sha256=e9kqHB6mhJgIIQV_Hv8QHg2P2R7dPVfd8Lt08Hi-UsQ,1374
 eva/vision/data/wsi/patching/samplers/base.py,sha256=KWLJMfaPk7-IZ-P2isYBvFAa5SuJPUhtD63hkKRFrgg,1287
 eva/vision/data/wsi/patching/samplers/foreground_grid.py,sha256=EhXkr5EFz2-RXEisWtjDa4CUTnrW4fiamQjEgALB2aI,3093
 eva/vision/data/wsi/patching/samplers/grid.py,sha256=dImrMSyCL3E_j5KRqpVJUWTe-mrJpfttg1Z9rbm3j0k,1363
-eva/vision/data/wsi/patching/samplers/random.py,sha256=qx5vExkmLgMFZgEwaXMmYFxoS-ewBhX-1Bpb1GGYkuI,1151
-eva/vision/losses/__init__.py,sha256=ZfUHa7siD3bBjiG4f39Eh4A0auaz0ctIKK0M9qfI-gY,95
-eva/vision/losses/dice.py,sha256=_D8Cj_m9AbOUhJS-GfsBbhfC-R9J58ao8UmuV_6OMhI,1424
+eva/vision/data/wsi/patching/samplers/random.py,sha256=0clmwCZ47bnTaSFke7jtjsmrFoY1ID2LjoiaE52dC3o,1228
+eva/vision/losses/__init__.py,sha256=htafabZgVcqbJjPURwsmGJ7AT6hIXc1-9SEuuaGU9SA,121
+eva/vision/losses/dice.py,sha256=qIMxtQlBbzES3hJ7x7pq0bd0GvIFpfGNL3KnMi0vRds,3669
+eva/vision/metrics/__init__.py,sha256=NtbcCrAUhVMMxSygTnbvNceJZBUzUD2tZp4nDmFjG3w,360
+eva/vision/metrics/defaults/__init__.py,sha256=ncQ9uH5q5SpfalyPX6dINPRLk34HLw6z9u8ny_HHbFQ,174
+eva/vision/metrics/defaults/segmentation/__init__.py,sha256=ve6dwyfhJGYBYKS6l6OySCBs32JnEBFnvhAyNvj-Uqo,191
+eva/vision/metrics/defaults/segmentation/multiclass.py,sha256=Qk4-OC0oujg99MEwRIjif-fS2mbAiWSFTZVRq4cY_Vo,1758
+eva/vision/metrics/segmentation/BUILD,sha256=Nf7BYWWe1USoFEIsIiEVZ8sa05J5FPkMJ-UIMDLrU8o,17
+eva/vision/metrics/segmentation/__init__.py,sha256=lfuyjuo2XjS7_dvaL-8e_LhLsyK2mc_mNGwyGqcb5X4,234
+eva/vision/metrics/segmentation/_utils.py,sha256=ebxTqymtxZ0iwMiH2snQHV_NVfDhZUSBlBS9AShWu_8,2464
+eva/vision/metrics/segmentation/generalized_dice.py,sha256=FqFzo7YWBwSlihmlgQg-O_ld1ZBQma0YTXk5XZ7faZM,2443
+eva/vision/metrics/segmentation/mean_iou.py,sha256=xR3wQOHT77SNKTRRPdDaWpJ88qgk9PIBT5n2lnKTUfM,2161
 eva/vision/models/__init__.py,sha256=a-P6JL73A3miHQnqgqUz07XtVmQB_o4DqPImk5rEATo,275
 eva/vision/models/modules/__init__.py,sha256=vaM_V6OF2s0lYjralP8dzv8mAtv_xIMZItfXgz0NZg8,156
-eva/vision/models/modules/semantic_segmentation.py,sha256=poBss37CM-bGLrtAl08WTcJtQgzwEP1MJgjeEbxexk0,6255
+eva/vision/models/modules/semantic_segmentation.py,sha256=VrzQemVgJbbXVQVzvvDjRaeXNhHsWDXRmTXGZhe-VCo,6389
 eva/vision/models/networks/__init__.py,sha256=j43IurizNlAyKPH2jwDHaeq49L2QvwbHWqUaptA1mG4,100
 eva/vision/models/networks/abmil.py,sha256=N1eH4fn1nXmgXurSQyQIxxonv7nsqeeuPWaQSHeltfs,6796
 eva/vision/models/networks/backbones/__init__.py,sha256=LsMx92eEoCQ5aNVFp7mHjrD-9ZeNawMiK6zZSYzl_PU,296
-eva/vision/models/networks/backbones/_utils.py,sha256=I8YrBsIVtCsp13xs1ln_OrhKBRu2gOmJdopL9hx_MBk,1277
-eva/vision/models/networks/backbones/pathology/__init__.py,sha256=-kn7JCC7fs8-VvjGQURQsdQejKYOwhPJch37Cf1crDM,1005
+eva/vision/models/networks/backbones/_utils.py,sha256=V7xeod4mElEuuO1TRW0xJE051cUyS1Saraw3-KcK1Mw,1667
+eva/vision/models/networks/backbones/pathology/__init__.py,sha256=goR59h8bfzd-Wa3rxPPdaSlAOH_df8SHBkTSKi08TS8,1147
 eva/vision/models/networks/backbones/pathology/bioptimus.py,sha256=wUSKjYgxcRV3FRHGaPwF1uRAQcGO0rHNHGmK1QDJXk4,991
 eva/vision/models/networks/backbones/pathology/gigapath.py,sha256=mfGXtKhY7XLpKQQAFNVZYsM-aeHCEbOVUrxpAEOr-l8,955
-eva/vision/models/networks/backbones/pathology/histai.py,sha256=C05W_75bINtTnet25M0axiVt00TMmcCx2U5Fcr7n-_I,1570
+eva/vision/models/networks/backbones/pathology/histai.py,sha256=X_we3U7GK91RrXyOX2PJB-YFDF2ozdL2fzZhNxm9SVU,1914
 eva/vision/models/networks/backbones/pathology/kaiko.py,sha256=GSdBG4WXrs1PWB2hr-sy_dFe2riwpPKwHx71esDoVfE,3952
 eva/vision/models/networks/backbones/pathology/lunit.py,sha256=ku4lr9pWeeHatHN4x4OVgwlve9sVqiRqIbgI0PXLiqg,2160
-eva/vision/models/networks/backbones/pathology/mahmood.py,sha256=3iIGKD7AvPDTritNkT2NGd6Nb5iJQxBKPmymI5YpOzo,2042
-eva/vision/models/networks/backbones/pathology/owkin.py,sha256=EdP4d1ndIR4URZHuSVfn01AdSe5n77KB5Sq4XEFmYwo,713
+eva/vision/models/networks/backbones/pathology/mahmood.py,sha256=me8DXf9nsEegDmltP8f7ZnG89xYVEKzZLKfVzMZjWDs,1832
+eva/vision/models/networks/backbones/pathology/owkin.py,sha256=uWJV5fgY7UZX6ilgGzkPY9fnlOiF03W7E8rc9TmlHGg,1231
+eva/vision/models/networks/backbones/pathology/paige.py,sha256=MjOLgdEKk8tdAIpCiHelasGwPE7xgzaooW6EE7IsuEE,1642
 eva/vision/models/networks/backbones/registry.py,sha256=anjILtEHHB6Ltwiw22h1bsgWtIjh_l5_fkPh87K7-d0,1631
 eva/vision/models/networks/backbones/timm/__init__.py,sha256=cZH3av9gIZcvEVD0rwKsI-MEq7zPqaW4dQ0E05CksvQ,128
 eva/vision/models/networks/backbones/timm/backbones.py,sha256=fCTiwqU6NhQ-ccAMzmpPDddXkFzRAB3mw4lcQ9um_PU,1646
 eva/vision/models/networks/backbones/universal/__init__.py,sha256=MAlkALSJ2_w6spSbB7NmKlL0Jsk1YKEycatdI0xO0_I,252
 eva/vision/models/networks/backbones/universal/vit.py,sha256=kpUCoXpefR34hRNlQDFK9lGr4oqS8Mn5vTLKWZ-gaOs,1820
-eva/vision/models/networks/decoders/__init__.py,sha256=kW79anaDHRm0Tkxt7ZIpYpaMggx8RGK2mogs77n-c6k,190
-eva/vision/models/networks/decoders/decoder.py,sha256=0tEx-eWEbNA53oafUbJkTb3j0watPpdntXMrQ66azsU,150
-eva/vision/models/networks/decoders/segmentation/__init__.py,sha256=bdGL_R44cyutqNXEMYMwA_RtqbdTL5xt2TVdS5BjGps,439
-eva/vision/models/networks/decoders/segmentation/common.py,sha256=4gxTimvc-JRzbIlD4yfGWXIjcEJSP_iY79h-mheDryc,2525
-eva/vision/models/networks/decoders/segmentation/conv2d.py,sha256=fv-0tF7_Ey4EH5iW08enPoaRrziiqbCfjrl1i50ZgfI,4092
-eva/vision/models/networks/decoders/segmentation/linear.py,sha256=89kDvs-e7Y3Bs3TQvmt2K7_cQYkv0T65A_nBh_anqFQ,4736
+eva/vision/models/networks/decoders/__init__.py,sha256=RXFWmoYw2i6E9VOUCJmU8c72icHannVuo-cUKy6fnLM,200
+eva/vision/models/networks/decoders/segmentation/__init__.py,sha256=N6jrhXHj0P7i7RptZbZ-JFehT2BM7meFyNIK0owAkaE,517
+eva/vision/models/networks/decoders/segmentation/base.py,sha256=b2TIJKiJR9vejVRpNyedMJLPTrpHhAEXvco8atb9TPU,411
+eva/vision/models/networks/decoders/segmentation/decoder2d.py,sha256=0jZrgFSdH5nlMYlbBmDb1E4kIQ3cG-LNOsiij51_NSA,4447
+eva/vision/models/networks/decoders/segmentation/linear.py,sha256=-i9RVaKM1UsB3AXDDKdMmHiD7y2sr5HfF-WvkB47Fhw,4743
+eva/vision/models/networks/decoders/segmentation/semantic/__init__.py,sha256=Ubs8GXyQpEHs26JUeUuiVP3jfn47eiBZM_UVbu749XU,398
+eva/vision/models/networks/decoders/segmentation/semantic/common.py,sha256=fPTb0T-2FiOU-jT81ynASKaW7fJiRk6vQjuPkzHOluc,2530
+eva/vision/models/networks/decoders/segmentation/semantic/with_image.py,sha256=I5PyGKKo8DcXYcw4xlCFzuavRJNRrzGT-szpDidMPXI,3516
+eva/vision/models/networks/decoders/segmentation/typings.py,sha256=8zAqIJLlQdCjsx-Dl4lnF4BB1VxTg_AyIquBVwpZlHg,537
 eva/vision/models/wrappers/__init__.py,sha256=8MT8qFM4nUXGpK1_i3rp70ODkOjn2KhhRo2I17qZCPM,210
 eva/vision/models/wrappers/from_registry.py,sha256=gdnxyg9drqlxfTNuS3aLbWGbZIwX1VNl0uudfjzVsXM,1614
 eva/vision/models/wrappers/from_timm.py,sha256=Z38Nb1i6OPKkgvFZOvGx-O3AZQuscf1zRVyrEBXQdJg,2320
 eva/vision/utils/__init__.py,sha256=vaUovprE743SmyFH8l6uk4pYSWpI4zxn7lN0EwePTJI,96
 eva/vision/utils/colormap.py,sha256=P904auPzaxGESTjFcbv550fc49DeXklSHkuhXWFXCEo,2384
 eva/vision/utils/convert.py,sha256=fqGmKrg5-JJLrTkTXB4YDcWTudXPrO1gGjsckVRUesU,1881
-eva/vision/utils/io/__init__.py,sha256=B9z6YiPUTI2aNDvN7t90_WugPE-L1d_1017aNeOkuZo,517
+eva/vision/utils/io/__init__.py,sha256=XGJ_W94DVEYXJ_tVpr_20NMpR5JLWEWHGF3v9Low79A,610
 eva/vision/utils/io/_utils.py,sha256=JzOt7Frj6ScF_aNjFtfHBn4ROnl6NhUZucmQhLc4Cww,768
 eva/vision/utils/io/image.py,sha256=IdOkr5MYqhYHz8U9drZ7wULTM3YHwCWSjZlu_Qdl4GQ,2053
 eva/vision/utils/io/mat.py,sha256=qpGifyjmpE0Xhv567Si7-zxKrgkgE0sywP70cHiLFGU,808
-eva/vision/utils/io/nifti.py,sha256=O_5x3A7RySfZYkF8KG5nmLQf1FcbhnJBVNVf71m3Lo4,2189
+eva/vision/utils/io/nifti.py,sha256=Q8Cd-ovqGZbevqfhb4waS6xI5xV3DXoWnDd5rhzLRNU,2595
 eva/vision/utils/io/text.py,sha256=qYgfo_ZaDZWfG02NkVVYzo5QFySqdCCz5uLA9d-zXtI,701
-kaiko_eva-0.1.1.dist-info/METADATA,sha256=Hk2Di7kFekY8Pneo4H7oE0lSsUBkI05hIQM_1Tby0qE,26806
-kaiko_eva-0.1.1.dist-info/WHEEL,sha256=Vza3XR51HW1KmFP0iIMUVYIvz0uQuKJpIXKYOBGQyFQ,90
-kaiko_eva-0.1.1.dist-info/entry_points.txt,sha256=6CSLu9bmQYJSXEg8gbOzRhxH0AGs75BB-vPm3VvfcNE,88
-kaiko_eva-0.1.1.dist-info/licenses/LICENSE,sha256=e6AEzr7j_R-PYr2qLO-JwLn8y70jbVD3U2mxbRmwcI4,11338
-kaiko_eva-0.1.1.dist-info/RECORD,,
+kaiko_eva-0.1.3.dist-info/METADATA,sha256=xgKnK4lR6GSdWW0oB52wY7spKYlq_jq19AAjJREHpBg,24869
+kaiko_eva-0.1.3.dist-info/WHEEL,sha256=thaaA2w1JzcGC48WYufAs8nrYZjJm8LqNfnXFOFyCC4,90
+kaiko_eva-0.1.3.dist-info/entry_points.txt,sha256=6CSLu9bmQYJSXEg8gbOzRhxH0AGs75BB-vPm3VvfcNE,88
+kaiko_eva-0.1.3.dist-info/licenses/LICENSE,sha256=e6AEzr7j_R-PYr2qLO-JwLn8y70jbVD3U2mxbRmwcI4,11338
+kaiko_eva-0.1.3.dist-info/RECORD,,

{kaiko_eva-0.1.1.dist-info → kaiko_eva-0.1.3.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: pdm-backend (2.4.1)
+Generator: pdm-backend (2.4.3)
 Root-Is-Purelib: true
 Tag: py3-none-any

eva/core/metrics/mean_iou.py

@@ -1,120 +0,0 @@
-"""Mean Intersection over Union (mIoU) metric for semantic segmentation."""
-
-from typing import Any, Literal, Tuple
-
-import torch
-import torchmetrics
-
-
-class MeanIoU(torchmetrics.Metric):
-    """Computes Mean Intersection over Union (mIoU) for semantic segmentation.
-
-    Fixes the torchmetrics implementation
-    (issue https://github.com/Lightning-AI/torchmetrics/issues/2558)
-    """
-
-    def __init__(
-        self,
-        num_classes: int,
-        include_background: bool = True,
-        ignore_index: int | None = None,
-        per_class: bool = False,
-        **kwargs: Any,
-    ) -> None:
-        """Initializes the metric.
-
-        Args:
-            num_classes: The number of classes in the segmentation problem.
-            include_background: Whether to include the background class in the computation
-            ignore_index: Integer specifying a target class to ignore. If given, this class
-                index does not contribute to the returned score, regardless of reduction method.
-            per_class: Whether to compute the IoU for each class separately. If set to ``False``,
-                the metric will compute the mean IoU over all classes.
-            kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.
-        """
-        super().__init__(**kwargs)
-
-        self.num_classes = num_classes
-        self.include_background = include_background
-        self.ignore_index = ignore_index
-        self.per_class = per_class
-
-        self.add_state("intersection", default=torch.zeros(num_classes), dist_reduce_fx="sum")
-        self.add_state("union", default=torch.zeros(num_classes), dist_reduce_fx="sum")
-
-    def update(self, preds: torch.Tensor, target: torch.Tensor) -> None:
-        """Update the state with the new data."""
-        intersection, union = _compute_intersection_and_union(
-            preds,
-            target,
-            num_classes=self.num_classes,
-            include_background=self.include_background,
-            ignore_index=self.ignore_index,
-        )
-        self.intersection += intersection.sum(0)
-        self.union += union.sum(0)
-
-    def compute(self) -> torch.Tensor:
-        """Compute the final mean IoU score."""
-        iou_valid = torch.gt(self.union, 0)
-        iou = torch.where(
-            iou_valid,
-            torch.divide(self.intersection, self.union),
-            torch.nan,
-        )
-        if not self.per_class:
-            iou = torch.mean(iou[iou_valid])
-        return iou
-
-
-def _compute_intersection_and_union(
-    preds: torch.Tensor,
-    target: torch.Tensor,
-    num_classes: int,
-    include_background: bool = False,
-    input_format: Literal["one-hot", "index"] = "index",
-    ignore_index: int | None = None,
-) -> Tuple[torch.Tensor, torch.Tensor]:
-    """Compute the intersection and union for semantic segmentation tasks.
-
-    Args:
-        preds: Predicted tensor with shape (N, ...) where N is the batch size.
-            The shape can be (N, H, W) for 2D data or (N, D, H, W) for 3D data.
-        target: Ground truth tensor with the same shape as preds.
-        num_classes: Number of classes in the segmentation task.
-        include_background: Whether to include the background class in the computation.
-        input_format: Format of the input tensors.
-        ignore_index: Integer specifying a target class to ignore. If given, this class
-            index does not contribute to the returned score, regardless of reduction method.
-
-    Returns:
-        Two tensors representing the intersection and union for each class.
-        Shape of each tensor is (N, num_classes).
-
-    Note:
-        - If input_format is "index", the tensors are converted to one-hot encoding.
-        - If include_background is `False`, the background class
-          (assumed to be the first channel) is ignored in the computation.
-    """
-    if ignore_index is not None:
-        mask = target != ignore_index
-        mask = mask.all(dim=-1, keepdim=True)
-        preds = preds * mask
-        target = target * mask
-
-    if input_format == "index":
-        preds = torch.nn.functional.one_hot(preds, num_classes=num_classes)
-        target = torch.nn.functional.one_hot(target, num_classes=num_classes)
-
-    if not include_background:
-        preds[..., 0] = 0
-        target[..., 0] = 0
-
-    reduce_axis = list(range(1, preds.ndim - 1))
-
-    intersection = torch.sum(torch.logical_and(preds, target), dim=reduce_axis)
-    target_sum = torch.sum(target, dim=reduce_axis)
-    pred_sum = torch.sum(preds, dim=reduce_axis)
-    union = target_sum + pred_sum - intersection
-
-    return intersection, union

eva/vision/models/networks/decoders/decoder.py

@@ -1,7 +0,0 @@
-"""Semantic segmentation decoder base class."""
-
-from torch import nn
-
-
-class Decoder(nn.Module):
-    """Semantic segmentation decoder base class."""