Merge branch 'master' into 765-gdown-does-not-seem-to-be-working

.devcontainer/onCreateCommand.sh

@@ -5,4 +5,4 @@ pip install wheel
 pip install openvino-dev==2023.0.1 # [OPTIONAL] to generate optimized models for inference
 pip install mlcube_docker          # [OPTIONAL] to deploy GaNDLF models as MLCube-compliant Docker containers
 pip install medmnist==2.1.0
-pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
+pip install torch==2.1.0+cpu torchvision==0.16.0+cpu torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu

.devcontainer/postCreateCommand.sh

@@ -6,7 +6,7 @@
 # if runnning on a GPU machine, install the GPU version of pytorch
 if command -v nvidia-smi &> /dev/null
 then
-	pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
+	pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118
 fi
 
 pip install -e .

.github/workflows/devcontainer.yml

@@ -28,7 +28,7 @@ jobs:
     # if: steps.changed-files-specific.outputs.only_modified == 'false'
     - name: Detect and screen file changes
       id: changed-files-specific
-      uses: tj-actions/changed-files@v34
+      uses: tj-actions/changed-files@v41
       with:
           files: |
             docs/**

.github/workflows/docker-image.yml

@@ -25,8 +25,8 @@ jobs:
       fail-fast: false # So that remaining jobs don't instantly quit if one fails (e.g, CPU/ROCm don't upload if CUDA just fails to push to ghcr...)
       matrix:
         include: # Platform locates Dockerfile ("Dockerfile-{PLATFORM}"), docker tag has to be all lowercase alphanumeric for mysterious docker reasons
-          - platform: CUDA11.6
-            dockertag: cuda116
+          - platform: CUDA11.8
+            dockertag: cuda118
           - platform: CPU
             dockertag: cpu
           # - platform: ROCm

.github/workflows/mlcube-test.yml

@@ -32,7 +32,7 @@ jobs:
     # if: steps.changed-files-specific.outputs.only_modified == 'false'
     - name: Detect and screen file changes
       id: changed-files-specific
-      uses: tj-actions/changed-files@v34
+      uses: tj-actions/changed-files@v41
       with:
           files: |
             .github/*.md
@@ -73,7 +73,7 @@ jobs:
         python -m pip install --upgrade pip
         python -m pip install wheel
         python -m pip install openvino-dev==2023.0.1 mlcube_docker
-        pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
+        pip install torch==2.1.0+cpu torchvision==0.16.0+cpu torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu
         pip install -e .
     - name: Run mlcube deploy tests
       working-directory: ./testing

.github/workflows/openfl-test.yml

@@ -33,7 +33,7 @@ jobs:
     # if: steps.changed-files-specific.outputs.only_modified == 'false'
     - name: Detect and screen file changes
       id: changed-files-specific
-      uses: tj-actions/changed-files@v34
+      uses: tj-actions/changed-files@v41
       with:
           files: |
             .github/*.md
@@ -73,7 +73,7 @@ jobs:
         sudo apt-get install libvips libvips-tools -y
         python -m pip install --upgrade pip
         python -m pip install wheel
-        pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
+        pip install torch==2.1.0+cpu torchvision==0.16.0+cpu torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu
         pip install -e .
     - name: Run generic unit tests to download data and construct CSVs
       if: steps.changed-files-specific.outputs.only_modified == 'false' # Run on any non-docs change

.github/workflows/python-install-check.yml

@@ -0,0 +1,22 @@
+name: Install across python versions
+
+on: [push, pull_request]
+
+jobs:
+  pyversion_install:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10", "3.11"]
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          # all good if installation finishes
+          python -m pip install -e .

.github/workflows/python-test.yml

@@ -33,7 +33,7 @@ jobs:
     # if: steps.changed-files-specific.outputs.only_modified == 'false'
     - name: Detect and screen file changes
       id: changed-files-specific
-      uses: tj-actions/changed-files@v34
+      uses: tj-actions/changed-files@v41
       with:
           files: |
             .github/*.md
@@ -74,7 +74,7 @@ jobs:
         python -m pip install --upgrade pip
         python -m pip install wheel
         python -m pip install openvino-dev==2023.0.1 mlcube_docker
-        pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
+        pip install torch==2.1.0+cpu torchvision==0.16.0+cpu torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu
         pip install -e .
     - name: Run generic unit tests
       if: steps.changed-files-specific.outputs.only_modified == 'false' # Run on any non-docs change

Dockerfile-CPU

@@ -9,7 +9,7 @@ RUN add-apt-repository ppa:deadsnakes/ppa
 RUN apt-get update &&  apt-get install -y  python3.9 python3-pip libjpeg8-dev zlib1g-dev python3-dev libpython3.9-dev libffi-dev libgl1
 RUN python3.9 -m pip install --upgrade pip
 # EXPLICITLY install cpu versions of torch/torchvision (not all versions have +cpu modes on PyPI...)
-RUN python3.9 -m pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
+RUN python3.9 -m pip install torch==2.1.0+cpu torchvision==0.16.0+cpu torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cpu
 RUN python3.9 -m pip install openvino-dev==2023.0.1 opencv-python-headless mlcube_docker
 
 # Do some dependency installation separately here to make layer caching more efficient

Dockerfile-CUDA11.6 → Dockerfile-CUDA11.8

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:11.6.2-devel-ubuntu20.04
+FROM nvidia/cuda:11.8.0-devel-ubuntu20.04
 LABEL github="https://github.com/mlcommons/GaNDLF"
 LABEL docs="https://mlcommons.github.io/GaNDLF/"
 LABEL version=1.0
@@ -12,7 +12,7 @@ RUN apt-get update && apt-get install -y software-properties-common
 RUN add-apt-repository ppa:deadsnakes/ppa
 RUN apt-get update &&  apt-get install -y  python3.9 python3-pip libjpeg8-dev zlib1g-dev python3-dev libpython3.9-dev libffi-dev libgl1
 RUN python3.9 -m pip install --upgrade pip
-RUN python3.9 -m pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
+RUN python3.9 -m pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118
 RUN python3.9 -m pip install openvino-dev==2023.0.1 opencv-python-headless mlcube_docker
 
 # Do some dependency installation separately here to make layer caching more efficient

Dockerfile-ROCm

@@ -1,4 +1,4 @@
-FROM rocm/pytorch:rocm5.2.3_ubuntu20.04_py3.7_pytorch_1.12.1
+FROM rocm/pytorch:rocm5.6_ubuntu20.04_py3.8_pytorch_1.12.1
 LABEL github="https://github.com/mlcommons/GaNDLF"
 LABEL docs="https://mlcommons.github.io/GaNDLF/"
 LABEL version=1.0
@@ -10,7 +10,7 @@ RUN apt-get update && apt-get install -y software-properties-common
 RUN add-apt-repository ppa:deadsnakes/ppa
 RUN apt-get update &&  apt-get install -y  python3.9 python3-pip libjpeg8-dev zlib1g-dev python3-dev libpython3.9-dev libffi-dev libgl1
 RUN python3.9 -m pip install --upgrade pip
-RUN python3.9 -m pip install torch==1.13.1+rocm5.2 torchvision==0.14.1+rocm5.2 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/rocm5.2
+RUN python3.9 -m pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/rocm5.6
 RUN python3.9 -m pip install --upgrade pip && python3.9 -m pip install openvino-dev==2023.0.1 opencv-python-headless mlcube_docker
 RUN apt-get update && apt-get install -y libgl1
 

GANDLF/cli/__init__.py

@@ -13,5 +13,5 @@
     "Contact: [email protected]\n\n"
     + "This program is NOT FDA/CE approved and NOT intended for clinical use.\nCopyright (c) "
     + str(date.today().year)
-    + " MLCommons. All rights reserved.\n\nCitation: https://arxiv.org/abs/2103.01006"
+    + " MLCommons. All rights reserved.\n\nCitation: https://doi.org/10.1038/s44172-023-00066-3"
 )

GANDLF/cli/preprocess_and_save.py

@@ -105,15 +105,13 @@ def preprocess_and_save(
             # the "squeeze" is needed because the dataloader automatically
             # constructs 5D tensor considering the batch_size as first
             # dimension, but the constructor needs 4D tensor.
-            subject_process[str(channel)] = torchio.Image(
+            subject_process[str(channel)] = torchio.ScalarImage(
                 tensor=subject[str(channel)]["data"].squeeze(0),
-                type=torchio.INTENSITY,
                 path=subject[str(channel)]["path"],
             )
         if parameters["headers"]["labelHeader"] is not None:
-            subject_process["label"] = torchio.Image(
+            subject_process["label"] = torchio.LabelMap(
                 tensor=subject["label"]["data"].squeeze(0),
-                type=torchio.LABEL,
                 path=subject["label"]["path"],
             )
         subject_dict_to_write = torchio.Subject(subject_process)

GANDLF/compute/forward_pass.py

@@ -129,12 +129,12 @@ def validate_network(
         subject_dict = {}
         label_ground_truth = None
         label_present = False
-        # this is when we want the dataloader to pick up properties of GaNDLF's DataLoader, such as pre-processing and augmentations, if appropriate
+        # this is when we want the dataloader to pick up properties of GaNDLF's
+        # DataLoader, such as pre-processing and augmentations, if appropriate
         if "label" in subject:
             if subject["label"] != ["NA"]:
-                subject_dict["label"] = torchio.Image(
+                subject_dict["label"] = torchio.LabelMap(
                     path=subject["label"]["path"],
-                    type=torchio.LABEL,
                     tensor=subject["label"]["data"].squeeze(0),
                     affine=subject["label"]["affine"].squeeze(0),
                 )
@@ -149,9 +149,8 @@ def validate_network(
                 )
 
         for key in params["channel_keys"]:
-            subject_dict[key] = torchio.Image(
+            subject_dict[key] = torchio.ScalarImage(
                 path=subject[key]["path"],
-                type=subject[key]["type"],
                 tensor=subject[key]["data"].squeeze(0),
                 affine=subject[key]["affine"].squeeze(0),
             )
@@ -320,8 +319,7 @@ def validate_network(
                 output_prediction = aggregator.get_output_tensor()
                 output_prediction = output_prediction.unsqueeze(0)
                 if params["save_output"]:
-                    img_for_metadata = torchio.Image(
-                        type=subject["1"]["type"],
+                    img_for_metadata = torchio.ScalarImage(
                         tensor=subject["1"]["data"].squeeze(0),
                         affine=subject["1"]["affine"].squeeze(0),
                     ).as_sitk()

GANDLF/data/augmentation/noise_enhanced.py

@@ -59,8 +59,8 @@ def get_params(
         mean_range: Tuple[float, float],
         std_range: Tuple[float, float],
     ) -> Tuple[float, float]:
-        mean = self.sample_uniform(*mean_range).item()
-        std = self.sample_uniform(*std_range).item()
+        mean = self.sample_uniform(*mean_range)
+        std = self.sample_uniform(*std_range)
         seed = self._get_random_seed()
         return mean, std, seed
 

GANDLF/data/patch_miner/opm/utils.py

@@ -10,8 +10,9 @@
 from skimage.morphology import remove_small_holes
 from skimage.color.colorconv import rgb2hsv
 import cv2
-#from skimage.exposure import rescale_intensity
-#from skimage.color import rgb2hed
+
+# from skimage.exposure import rescale_intensity
+# from skimage.color import rgb2hed
 
 # import matplotlib.pyplot as plt
 import yaml
@@ -238,7 +239,8 @@ def alpha_rgb_2d_channel_check(img):
     else:
         return False
 
-#def pen_marking_check(img, intensity_thresh=225, intensity_thresh_saturation =50, intensity_thresh_b = 128):
+
+# def pen_marking_check(img, intensity_thresh=225, intensity_thresh_saturation =50, intensity_thresh_b = 128):
 #    """
 #    This function is used to curate patches from the input image. It is used to remove patches that have pen markings.
 #    Args:
@@ -259,7 +261,14 @@ def alpha_rgb_2d_channel_check(img):
 #    #Assume patch is valid
 #    return True
 
-def patch_artifact_check(img, intensity_thresh = 250, intensity_thresh_saturation = 5, intensity_thresh_b = 128, patch_size = (256,256)):
+
+def patch_artifact_check(
+    img,
+    intensity_thresh=250,
+    intensity_thresh_saturation=5,
+    intensity_thresh_b=128,
+    patch_size=(256, 256),
+):
     """
     This function is used to curate patches from the input image. It is used to remove patches that are mostly background.
     Args:
@@ -271,23 +280,36 @@ def patch_artifact_check(img, intensity_thresh = 250, intensity_thresh_saturatio
     Returns:
         bool: Whether the patch is valid (True) or not (False)
     """
-    #patch_size = config["patch_size"]
+    # patch_size = config["patch_size"]
     patch_hsv = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
     count_white_pixels = np.sum(np.logical_and.reduce(img > intensity_thresh, axis=2))
     percent_pixels = count_white_pixels / (patch_size[0] * patch_size[1])
     count_black_pixels = np.sum(np.logical_and.reduce(img < intensity_thresh_b, axis=2))
     percent_pixel_b = count_black_pixels / (patch_size[0] * patch_size[1])
-    percent_pixel_2 = np.sum(patch_hsv[...,1] < intensity_thresh_saturation) / (patch_size[0] * patch_size[1])
-    percent_pixel_3 = np.sum(patch_hsv[...,2] > intensity_thresh) / (patch_size[0] * patch_size[1])
+    percent_pixel_2 = np.sum(patch_hsv[..., 1] < intensity_thresh_saturation) / (
+        patch_size[0] * patch_size[1]
+    )
+    percent_pixel_3 = np.sum(patch_hsv[..., 2] > intensity_thresh) / (
+        patch_size[0] * patch_size[1]
+    )
 
-    if percent_pixel_2 > 0.99 or np.mean(patch_hsv[...,1]) < 5 or percent_pixel_3 > 0.99:
+    if (
+        percent_pixel_2 > 0.99
+        or np.mean(patch_hsv[..., 1]) < 5
+        or percent_pixel_3 > 0.99
+    ):
         if percent_pixel_2 < 0.1:
             return False
-    elif (percent_pixel_2 > 0.99 and percent_pixel_3 > 0.99) or percent_pixel_b > 0.99 or percent_pixels > 0.9:
+    elif (
+        (percent_pixel_2 > 0.99 and percent_pixel_3 > 0.99)
+        or percent_pixel_b > 0.99
+        or percent_pixels > 0.9
+    ):
         return False
     # assume that the patch is valid
     return True
 
+
 def parse_config(config_file):
     """
     Parse config file and return a dictionary of config values.
@@ -304,7 +326,7 @@ def parse_config(config_file):
     config["value_map"] = config.get("value_map", None)
     config["read_type"] = config.get("read_type", "random")
     config["overlap_factor"] = config.get("overlap_factor", 0.0)
-    config["patch_size"] = config.get("patch_size", [256,256])
+    config["patch_size"] = config.get("patch_size", [256, 256])
 
     return config
 

GANDLF/metrics/classification.py

@@ -64,9 +64,7 @@ def overall_stats(predictions, ground_truth, params):
             "aucroc": tm.AUROC(
                 task=task,
                 num_classes=params["model"]["num_classes"],
-                average=average_type_key
-                if average_type_key != "micro"
-                else "macro",
+                average=average_type_key if average_type_key != "micro" else "macro",
             ),
         }
         for metric_name, calculator in calculators.items():

GANDLF/metrics/generic.py

@@ -15,9 +15,7 @@
 )
 
 
-def generic_function_output_with_check(
-    predicted_classes, label, metric_function
-):
+def generic_function_output_with_check(predicted_classes, label, metric_function):
     if torch.min(predicted_classes) < 0:
         print(
             "WARNING: Negative values detected in prediction, cannot compute torchmetrics calculations."
@@ -32,16 +30,12 @@ def generic_function_output_with_check(
             max_clamp_val = metric_function.num_classes - 1
         except AttributeError:
             max_clamp_val = 1
-        predicted_new = torch.clamp(
-            predicted_classes.cpu().int(), max=max_clamp_val
-        )
+        predicted_new = torch.clamp(predicted_classes.cpu().int(), max=max_clamp_val)
         predicted_new = predicted_new.reshape(label.shape)
         return metric_function(predicted_new, label.cpu().int())
 
 
-def generic_torchmetrics_score(
-    output, label, metric_class, metric_key, params
-):
+def generic_torchmetrics_score(output, label, metric_class, metric_key, params):
     task = determine_classification_task_type(params)
     num_classes = params["model"]["num_classes"]
     predicted_classes = output
@@ -67,25 +61,19 @@ def recall_score(output, label, params):
 
 
 def precision_score(output, label, params):
-    return generic_torchmetrics_score(
-        output, label, Precision, "precision", params
-    )
+    return generic_torchmetrics_score(output, label, Precision, "precision", params)
 
 
 def f1_score(output, label, params):
     return generic_torchmetrics_score(output, label, F1Score, "f1", params)
 
 
 def accuracy(output, label, params):
-    return generic_torchmetrics_score(
-        output, label, Accuracy, "accuracy", params
-    )
+    return generic_torchmetrics_score(output, label, Accuracy, "accuracy", params)
 
 
 def specificity_score(output, label, params):
-    return generic_torchmetrics_score(
-        output, label, Specificity, "specificity", params
-    )
+    return generic_torchmetrics_score(output, label, Specificity, "specificity", params)
 
 
 def iou_score(output, label, params):