chore (CI): ensure support to torch 2.4 (kornia#2972)

.github/workflows/pr_test_cpu.yml

@@ -20,32 +20,32 @@ jobs:
         os: ['Ubuntu-latest', 'Windows-latest']
         pytorch-dtype: ['float32', 'float64']
 
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: ${{ matrix.os }}
       python-version: '["3.8", "3.12"]'
-      pytorch-version: '["1.9.1", "2.3.1"]'
+      pytorch-version: '["1.9.1", "2.4.0"]'
       pytorch-dtype: ${{ matrix.pytorch-dtype }}
 
   tests-cpu-macos:
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: 'MacOS-latest'
       python-version: '["3.8", "3.12"]'
       pytorch-dtype: 'float32'
 
 
   coverage:
-    uses: kornia/workflows/.github/workflows/coverage.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/coverage.yml@v1.11.0
 
   typing:
-    uses: kornia/workflows/.github/workflows/mypy.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/mypy.yml@v1.11.0
 
   tutorials:
-    uses: kornia/workflows/.github/workflows/tutorials.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tutorials.yml@v1.12.1
 
   docs:
-    uses: kornia/workflows/.github/workflows/docs.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/docs.yml@v1.11.0
     with:
       python-version: '["3.11"]'
 
@@ -68,7 +68,7 @@ jobs:
         os: ['Ubuntu-latest', 'Windows-latest'] #, 'MacOS-latest'] add it when https://github.com/pytorch/pytorch/pull/89262 be merged
         pytorch-dtype: ['float32', 'float64']
 
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: ${{ matrix.os }}
       pytorch-version: '["nightly"]'

.github/workflows/scheduled_test_cpu.yml

@@ -18,11 +18,11 @@ jobs:
         # os: ['Ubuntu-latest', 'Windows-latest', 'MacOS-latest']
         pytorch-dtype: ['float32', 'float64']
 
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: 'Ubuntu-latest'
       python-version: '["3.8", "3.9", "3.10", "3.11", "3.12"]'
-      pytorch-version: '["1.9.1", "1.10.2", "1.11.0", "1.12.1", "1.13.1", "2.0.1", "2.1.2", "2.2.2", "2.3.1"]'
+      pytorch-version: '["1.9.1", "1.10.2", "1.11.0", "1.12.1", "1.13.1", "2.0.1", "2.1.2", "2.2.2", "2.3.1", "2.4.0"]'
       pytorch-dtype: ${{ matrix.pytorch-dtype }}
       pytest-extra: '--runslow'
 
@@ -34,11 +34,11 @@ jobs:
       matrix:
         pytorch-dtype: ['float32', 'float64']
 
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: 'Windows-latest'
       python-version: '["3.12"]'
-      pytorch-version: '["1.9.1", "2.3.1"]'
+      pytorch-version: '["1.9.1", "2.4.0"]'
       pytorch-dtype: ${{ matrix.pytorch-dtype }}
 
   tests-cpu-mac:
@@ -47,21 +47,21 @@ jobs:
       matrix:
         pytorch-dtype: ['float32', 'float64']
 
-    uses: kornia/workflows/.github/workflows/tests.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tests.yml@v1.11.0
     with:
       os: 'MacOS-latest'
       pytorch-dtype: ${{ matrix.pytorch-dtype }}
 
   coverage:
-    uses: kornia/workflows/.github/workflows/coverage.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/coverage.yml@v1.11.0
 
   typing:
-    uses: kornia/workflows/.github/workflows/mypy.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/mypy.yml@v1.11.0
 
   tutorials:
-    uses: kornia/workflows/.github/workflows/tutorials.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/tutorials.yml@v1.12.1
 
   docs:
-    uses: kornia/workflows/.github/workflows/docs.yml@v1.10.1
+    uses: kornia/workflows/.github/workflows/docs.yml@v1.11.0
     with:
       python-version: '["3.11"]'

kornia/contrib/face_detection.py

@@ -8,7 +8,6 @@
 from torch import nn
 
 from kornia.geometry.bbox import nms as nms_kornia
-from kornia.utils.helpers import map_location_to_cpu
 
 __all__ = ["FaceDetector", "FaceDetectorResult", "FaceKeypoint"]
 
@@ -289,7 +288,7 @@ def __init__(self, phase: str, pretrained: bool) -> None:
 
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(url, map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(url, map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict, strict=True)
         self.eval()
 

kornia/feature/affine_shape.py

@@ -8,7 +8,6 @@
 from kornia.core.check import KORNIA_CHECK_LAF, KORNIA_CHECK_SHAPE
 from kornia.filters.kernels import get_gaussian_kernel2d
 from kornia.filters.sobel import SpatialGradient
-from kornia.utils.helpers import map_location_to_cpu
 
 from .laf import (
     ellipse_to_laf,
@@ -189,7 +188,7 @@ def __init__(self, pretrained: bool = False, preserve_orientation: bool = True)
         self.patch_size = 32
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["affnet"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["affnet"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict["state_dict"], strict=False)
         self.preserve_orientation = preserve_orientation
         if preserve_orientation:

kornia/feature/dedode/dedode.py

@@ -198,10 +198,10 @@ def from_pretrained(
             amp_dtype=amp_dtype,
         )
         model.detector.load_state_dict(
-            torch.hub.load_state_dict_from_url(urls["detector"][detector_weights], map_location=map_location_to_cpu)
+            torch.hub.load_state_dict_from_url(urls["detector"][detector_weights], map_location=torch.device("cpu"))
         )
         model.descriptor.load_state_dict(
-            torch.hub.load_state_dict_from_url(urls["descriptor"][descriptor_weights], map_location=map_location_to_cpu)
+            torch.hub.load_state_dict_from_url(urls["descriptor"][descriptor_weights], map_location=torch.device("cpu"))
         )
         model.eval()
         return model

kornia/feature/defmo.py

@@ -4,7 +4,6 @@
 from torch import nn
 
 from kornia.core import Module, Tensor, concatenate, stack
-from kornia.utils.helpers import map_location_to_cpu
 
 urls: Dict[str, str] = {}
 urls["defmo_encoder"] = "http://ptak.felk.cvut.cz/personal/rozumden/defmo_saved_models/encoder_best.pt"
@@ -288,11 +287,11 @@ def __init__(self, pretrained: bool = False) -> None:
         # use torch.hub to load pretrained model
         if pretrained:
             pretrained_dict = torch.hub.load_state_dict_from_url(
-                urls["defmo_encoder"], map_location=map_location_to_cpu
+                urls["defmo_encoder"], map_location=torch.device("cpu")
             )
             self.encoder.load_state_dict(pretrained_dict, strict=True)
             pretrained_dict_ren = torch.hub.load_state_dict_from_url(
-                urls["defmo_rendering"], map_location=map_location_to_cpu
+                urls["defmo_rendering"], map_location=torch.device("cpu")
             )
             self.rendering.load_state_dict(pretrained_dict_ren, strict=True)
         self.eval()

kornia/feature/hardnet.py

@@ -5,7 +5,7 @@
 from torch import nn
 
 from kornia.core.check import KORNIA_CHECK_SHAPE
-from kornia.utils.helpers import is_mps_tensor_safe, map_location_to_cpu
+from kornia.utils.helpers import is_mps_tensor_safe
 
 urls: Dict[str, str] = {}
 urls["hardnet++"] = "https://github.com/DagnyT/hardnet/raw/master/pretrained/pretrained_all_datasets/HardNet++.pth"
@@ -67,7 +67,7 @@ def __init__(self, pretrained: bool = False) -> None:
 
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty_aug"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty_aug"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict["state_dict"], strict=True)
         self.eval()
 
@@ -150,7 +150,7 @@ def __init__(self, pretrained: bool = False) -> None:
 
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["hardnet8v2"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["hardnet8v2"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict, strict=True)
         self.eval()
 

kornia/feature/hynet.py

@@ -4,7 +4,6 @@
 from torch import nn
 
 from kornia.core import Module, Parameter, Tensor, tensor, zeros
-from kornia.utils.helpers import map_location_to_cpu
 
 urls: Dict[str, str] = {}
 urls["liberty"] = "https://github.com/ducha-aiki/Key.Net-Pytorch/raw/main/model/HyNet/weights/HyNet_LIB.pth"  # pylint: disable
@@ -209,7 +208,7 @@ def __init__(
         self.desc_norm = nn.LocalResponseNorm(2 * self.dim_desc, 2.0 * self.dim_desc, 0.5, 0.0)
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict, strict=True)
         self.eval()
 

kornia/feature/keynet.py

@@ -8,7 +8,6 @@
 from kornia.core import Module, Tensor, concatenate
 from kornia.filters import SpatialGradient
 from kornia.geometry.transform import pyrdown
-from kornia.utils.helpers import map_location_to_cpu
 
 from .scale_space_detector import Detector_config, MultiResolutionDetector, get_default_detector_config
 
@@ -145,7 +144,7 @@ def __init__(self, pretrained: bool = False, keynet_conf: KeyNet_conf = keynet_d
         )
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(KeyNet_URL, map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(KeyNet_URL, map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict["state_dict"], strict=True)
         self.eval()
 

kornia/feature/loftr/loftr.py

@@ -6,7 +6,6 @@
 
 from kornia.core import Module, Tensor
 from kornia.geometry import resize
-from kornia.utils.helpers import map_location_to_cpu
 
 from .backbone import build_backbone
 from .loftr_module import FinePreprocess, LocalFeatureTransformer
@@ -97,7 +96,7 @@ def __init__(self, pretrained: Optional[str] = "outdoor", config: dict[str, Any]
             if pretrained not in urls.keys():
                 raise ValueError(f"pretrained should be None or one of {urls.keys()}")
 
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls[pretrained], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls[pretrained], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict["state_dict"])
         self.eval()
 

kornia/feature/mkd.py

@@ -9,7 +9,6 @@
 from kornia.filters import GaussianBlur2d, SpatialGradient
 from kornia.geometry.conversions import cart2pol
 from kornia.utils import create_meshgrid
-from kornia.utils.helpers import map_location_to_cpu
 
 # Precomputed coefficients for Von Mises kernel, given N and K(appa).
 sqrt2: float = 1.4142135623730951
@@ -534,7 +533,7 @@ def __init__(
         # Load supervised(lw)/unsupervised(pca) model trained on training_set.
         if self.whitening is not None:
             whitening_models = torch.hub.load_state_dict_from_url(
-                urls[self.kernel_type], map_location=map_location_to_cpu
+                urls[self.kernel_type], map_location=torch.device("cpu")
             )
             whitening_model = whitening_models[training_set]
             self.whitening_layer = Whitening(
@@ -582,7 +581,7 @@ def __repr__(self) -> str:
 
 
 def load_whitening_model(kernel_type: str, training_set: str) -> Dict[str, Any]:
-    whitening_models = torch.hub.load_state_dict_from_url(urls[kernel_type], map_location=map_location_to_cpu)
+    whitening_models = torch.hub.load_state_dict_from_url(urls[kernel_type], map_location=torch.device("cpu"))
     whitening_model = whitening_models[training_set]
     return whitening_model
 

kornia/feature/orientation.py

@@ -8,7 +8,6 @@
 from kornia.core.check import KORNIA_CHECK_LAF, KORNIA_CHECK_SHAPE
 from kornia.filters import SpatialGradient, get_gaussian_discrete_kernel1d, get_gaussian_kernel2d
 from kornia.geometry import rad2deg
-from kornia.utils.helpers import map_location_to_cpu
 
 from .laf import extract_patches_from_pyramid, get_laf_orientation, set_laf_orientation
 
@@ -162,7 +161,7 @@ def __init__(self, pretrained: bool = False, eps: float = 1e-8) -> None:
         self.eps = eps
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["orinet"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["orinet"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict["state_dict"], strict=False)
         self.eval()
 

kornia/feature/sold2/sold2.py

@@ -8,7 +8,7 @@
 from kornia.core.check import KORNIA_CHECK_SHAPE
 from kornia.feature.sold2.structures import DetectorCfg, LineMatcherCfg
 from kornia.geometry.conversions import normalize_pixel_coordinates
-from kornia.utils import dataclass_to_dict, dict_to_dataclass, map_location_to_cpu
+from kornia.utils import dataclass_to_dict, dict_to_dataclass
 
 from .backbones import SOLD2Net
 from .sold2_detector import LineSegmentDetectionModule, line_map_to_segments, prob_to_junctions
@@ -64,7 +64,7 @@ def __init__(self, pretrained: bool = True, config: Optional[DetectorCfg] = None
         # Load the pre-trained model
         self.model = SOLD2Net(dataclass_to_dict(self.config))
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["wireframe"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["wireframe"], map_location=torch.device("cpu"))
             state_dict = self.adapt_state_dict(pretrained_dict["model_state_dict"])
             self.model.load_state_dict(state_dict)
         self.eval()

kornia/feature/sold2/sold2_detector.py

@@ -8,7 +8,7 @@
 from kornia.core.check import KORNIA_CHECK_SHAPE
 from kornia.feature.sold2.structures import DetectorCfg, HeatMapRefineCfg, JunctionRefineCfg, LineDetectorCfg
 from kornia.geometry.bbox import nms
-from kornia.utils import dataclass_to_dict, dict_to_dataclass, map_location_to_cpu, torch_meshgrid
+from kornia.utils import dataclass_to_dict, dict_to_dataclass, torch_meshgrid
 
 from .backbones import SOLD2Net
 
@@ -58,7 +58,7 @@ def __init__(self, pretrained: bool = True, config: Optional[DetectorCfg] = None
         self.model = SOLD2Net(dataclass_to_dict(self.config))
 
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["wireframe"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["wireframe"], map_location=torch.device("cpu"))
             state_dict = self.adapt_state_dict(pretrained_dict["model_state_dict"])
             self.model.load_state_dict(state_dict)
         self.eval()

kornia/feature/sosnet.py

@@ -4,7 +4,6 @@
 from torch import nn
 
 from kornia.core.check import KORNIA_CHECK_SHAPE
-from kornia.utils.helpers import map_location_to_cpu
 
 urls: Dict[str, str] = {}
 urls["lib"] = "https://github.com/yuruntian/SOSNet/raw/master/sosnet-weights/sosnet_32x32_liberty.pth"
@@ -61,7 +60,7 @@ def __init__(self, pretrained: bool = False) -> None:
         self.desc_norm = nn.Sequential(nn.LocalResponseNorm(256, alpha=256.0, beta=0.5, k=0.0))
         # load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["lib"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["lib"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict, strict=True)
         self.eval()
 

kornia/feature/tfeat.py

@@ -4,7 +4,6 @@
 from torch import nn
 
 from kornia.core.check import KORNIA_CHECK_SHAPE
-from kornia.utils.helpers import map_location_to_cpu
 
 urls: Dict[str, str] = {}
 urls["liberty"] = "https://github.com/vbalnt/tfeat/raw/master/pretrained-models/tfeat-liberty.params"  # pylint: disable
@@ -50,7 +49,7 @@ def __init__(self, pretrained: bool = False) -> None:
         self.descr = nn.Sequential(nn.Linear(64 * 8 * 8, 128), nn.Tanh())
         # use torch.hub to load pretrained model
         if pretrained:
-            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty"], map_location=map_location_to_cpu)
+            pretrained_dict = torch.hub.load_state_dict_from_url(urls["liberty"], map_location=torch.device("cpu"))
             self.load_state_dict(pretrained_dict, strict=True)
         self.eval()
 

kornia/filters/dexined.py

@@ -12,7 +12,6 @@
 from kornia.core import ImageModule as Module
 from kornia.core import Tensor, concatenate
 from kornia.core.check import KORNIA_CHECK
-from kornia.utils import map_location_to_cpu
 
 url: str = "http://cmp.felk.cvut.cz/~mishkdmy/models/DexiNed_BIPED_10.pth"
 
@@ -222,7 +221,7 @@ def __init__(self, pretrained: bool) -> None:
 
     def load_from_file(self, path_file: str) -> None:
         # use torch.hub to load pretrained model
-        pretrained_dict = torch.hub.load_state_dict_from_url(path_file, map_location=map_location_to_cpu)
+        pretrained_dict = torch.hub.load_state_dict_from_url(path_file, map_location=torch.device("cpu"))
         self.load_state_dict(pretrained_dict, strict=True)
         self.eval()