[DRAFT] add tia toolbox

pyproject.toml

@@ -154,6 +154,7 @@ all = [
     "uvicorn[standard]>=0.32.0",
     "opencv-python-headless>=4.11.0.86",
     "histomicstk>=1.3.14",
+    "tiatoolbox>=2.1.3",
 ]
 
 dev = [
@@ -162,6 +163,9 @@ dev = [
 histomicstk = [
     "histomicstk>=1.3.14",
 ]
+tiatoolbox = [
+    "tiatoolbox>=2.1.3",
+]
 
 [build-system]
 requires = ["uv_build>=0.9.26,<0.10.0"]
@@ -171,8 +175,8 @@ build-backend = "uv_build"
 url = "https://girder.github.io/large_image_wheels"
 
 [[tool.uv.index]]
-name = "pytorch-cu129"
-url = "https://download.pytorch.org/whl/cu129"
+name = "pytorch-cu126"
+url = "https://download.pytorch.org/whl/cu126"
 explicit = true
 
 [tool.uv.build-backend]
@@ -185,11 +189,12 @@ trident = { git = "https://github.com/mahmoodlab/trident.git", rev = "adf3b7e8fd
 tissue-segmentation = { git = "https://github.com/imi-bigpicture/tissue-segmentation.git", subdirectory = "tissue_segmentation", rev = "6d97a25a8255f591eb2c705611a32a5f56101a25"}
 sam-2 = { git = "https://github.com/facebookresearch/sam2.git", rev = "2b90b9f5ceec907a1c18123530e92e794ad901a4" }
 torch = [
-  { index = "pytorch-cu129", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+  { index = "pytorch-cu126", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
 ]
 torchvision = [
-  { index = "pytorch-cu129", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+  { index = "pytorch-cu126", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
 ]
+tiatoolbox = { git = "https://github.com/TissueImageAnalytics/tiatoolbox", rev = "0482082c0934abf095b43cd60dd9b8db002755c5" }
 
 [tool.uv]
 index-strategy = "unsafe-best-match"
@@ -201,6 +206,9 @@ exclude-dependencies = [
     "tensorflow-cpu",
     "opencv-python",  # ultralytics installs the wrong opencv package
 ]
+override-dependencies = [
+    { package = { name = "trident", version = "0.2.3" }, dependencies = ["timm>=1.0.3"] },  # It's very strict on an early timm version, while tiatoolbox requires >=1.0.3.
+]
 
 [tool.uv.extra-build-dependencies]
 imagecodecs-numcodecs = ["imagecodecs", "numcodecs"]

run.py

@@ -24,6 +24,7 @@
 # =============================================================================
 
 METHODS = [
+    seg.TIAToolboxSegmenter(),
     seg.EntropyMaskerSegmenter(mpp=20),
     seg.BackgroundSubtractorMOG2Segmenter(mpp=20),
     seg.FESISegmenter(mpp=20),

segmenteer/__init__.py

@@ -128,6 +128,7 @@
     "TRIDENTHESTSegmenter": ("segmenteer.methods.dl.trident", "TRIDENTHESTSegmenter"),
     "TRIDENTPathProfilerSegmenter": ("segmenteer.methods.dl.trident", "TRIDENTPathProfilerSegmenter"),
     "TRIDENTCPGSegmenter": ("segmenteer.methods.dl.trident", "TRIDENTCPGSegmenter"),
+    "TIAToolboxSegmenter": ("segmenteer.methods.dl.tiatoolbox", "TIAToolboxSegmenter"),
 }
 
 

segmenteer/methods/dl/__init__.py

@@ -23,6 +23,11 @@
 except ImportError:
     RTLucassenSlideSegmenter = None  # type: ignore[assignment]
 
+try:
+    from segmenteer.methods.dl.tiatoolbox import TIAToolboxSegmenter
+except ImportError:
+    TIAToolboxSegmenter = None
+
 try:
     from segmenteer.methods.dl.trident import (
         TRIDENTGrandQCSegmenter,
@@ -46,4 +51,5 @@
     "BigPictureSegmenter",
     "TRIDENTPathProfilerSegmenter",
     "TRIDENTCPGSegmenter",
+    "TIAToolboxSegmenter",
 ]

segmenteer/methods/dl/tiatoolbox.py

@@ -0,0 +1,32 @@
+    # >>> from tiatoolbox.models.engine.semantic_segmentor import SemanticSegmentor
+    # >>> segmentor = SemanticSegmentor(model="efficientunet-tissue_mask")
+    # >>> results = segmentor.run(
+    # ...     ["/example_wsi.svs"],
+    # ...     masks=None,
+    # ...     auto_get_mask=False,
+    # ...     patch_mode=False,
+    # ...     save_dir=Path("/tissue_mask/"),
+    # ...     output_type="annotationstore",
+    # ... )
+
+from tiatoolbox.models.engine.semantic_segmentor import SemanticSegmentor
+
+from segmenteer.core.base import PathSegmenter
+
+class TIAToolboxSegmenter(PathSegmenter):
+    def name() -> str:
+        return "tiatoolbox"
+
+    def _segment_path(self, image_path, output_path):
+        segmentor = SemanticSegmentor(model="efficientunet-tissue_mask")
+        save_dir = output_path.parent / "output"
+        results = segmentor.run(
+            [image_path],
+            masks=None,
+            auto_get_mask=False,
+            patch_mode=False,
+            save_dir=save_dir,
+            output_type="dict",
+            device="cuda",
+        )
+        print(results)