Merge 295ca08 into e101ebe

.github/workflows/run_tests.yml

@@ -65,8 +65,8 @@ jobs:
       fail-fast: false
 
       matrix:
-        os: [ "macos-latest", "windows-latest", "ubuntu-20.04", "ubuntu-18.04" ]
-        python-version: [ '3.7', '3.8', '3.9', '3.10' ]
+        os: [ "macos-latest", "windows-latest", "ubuntu-20.04" ]
+        python-version: [ '3.8', '3.9', '3.10' ]
         test-name:
           - integration-test
 

.github/workflows/run_tests_dummy.yml

@@ -28,7 +28,7 @@ jobs:
     strategy:
       matrix:
         # This list must be kept **in sync** with the Required Statuses in https://github.com/ivadomed/ivadomed/settings/branch_protection_rules/5051948
-        os: [ "macos-latest", "windows-latest", "ubuntu-20.04", "ubuntu-18.04" ]
+        os: [ "macos-latest", "ubuntu-20.04" ]
         python-version: [ 3.8 ]
         test-name:
           - integration-test

.readthedocs.yml

@@ -10,7 +10,7 @@ sphinx:
   configuration: docs/source/conf.py
 
 python:
-  version: 3.7
+  version: 3.10
   install:
     # Install IvadoMed package via setup.py content
     - method: pip

ivadomed/inference.py

@@ -225,7 +225,8 @@ def pred_to_png(pred_list: list, target_list: list, subj_path: str, suffix: str
     for pred, target in zip(pred_list, target_list):
         filename = subj_path + target + suffix
         data = pred.get_fdata()
-        imageio.imwrite(filename, (data*255/max_value).astype(np.uint8), format='png')
+        _img = (data*255/max_value).astype(np.uint8).squeeze()
+        imageio.v3.imwrite(filename, _img, extension='.png')
 
 
 def process_transformations(context: dict, fname_roi: str, fname_prior: str, metadata: dict, slice_axis: int,

ivadomed/loader/segmentation_pair.py

@@ -315,12 +315,28 @@ def convert_file_to_nifti(self, filename: str, extension: str, is_gt: bool = Fal
         """
         # For '.png', '.tif', '.tiff', '.jpg' and 'jpeg' extentions
         # Read image as 8 bit grayscale in numpy array (behavior TBD in ivadomed for RGB, RBGA or higher bit depth)
-        if "tif" in extension:
-            img = np.expand_dims(imageio.v2.imread(filename, format='tiff-pil'), axis=-1).astype(np.uint8)
-            if len(img.shape) > 3:
-                img = np.expand_dims(imageio.v2.imread(filename, format='tiff-pil', as_gray=True), axis=-1).astype(np.uint8)
-        else:
-            img = np.expand_dims(imageio.v2.imread(filename, as_gray=True), axis=-1).astype(np.uint8)
+        props = imageio.v3.improps(filename) # new in v3 - standardized metadata
+        _img = imageio.v3.imread(filename)
+
+        # TIFF is a "wild" format. A few assumptions greatly simplify the code below:
+        # 1. the image is interleaved/channel-last (not planar)
+        # 2. the colorspace is one of: binary, gray, RGB, RGBA (not aliasing ones like YUV or CMYK)
+        # 3. the image is flat (not a volume or time-series)
+        # Note: All of these are trivially true for JPEG and PNG due to limitations of these formats.
+
+        # make grayscale (treats binary as 1-bit grayscale)
+        colorspace_idx = 2 + int(props.is_batch)
+        if _img.ndim <= colorspace_idx:  # binary or gray
+            pass  # nothing to do
+        elif _img.shape[colorspace_idx] == 3:  # RGB
+            _img = np.sum(_img * (.299, .587, .114), axis=-1)
+        else:  # RGBA
+            # discards alpha
+            _img = np.sum(_img * (.299, .587, .114, 0), axis=-1)
+        if len(_img.shape) < 3:
+            _img = np.expand_dims(_img, axis=-1)
+
+        img = imageio.core.image_as_uint(_img, bitdepth=8)
 
         # Binarize ground-truth values (0-255) to 0 and 1 in uint8 with threshold 0.5
         if is_gt:

ivadomed/loader/utils.py

@@ -13,7 +13,6 @@
 import torch
 from loguru import logger
 from sklearn.model_selection import train_test_split
-from torch._six import string_classes
 from ivadomed import utils as imed_utils
 from ivadomed.keywords import SplitDatasetKW, LoaderParamsKW, ROIParamsKW, ContrastParamsKW
 import nibabel as nib
@@ -270,7 +269,7 @@ def imed_collate(batch: dict) -> dict | list | str | torch.Tensor:
         return torch.LongTensor(batch)
     elif isinstance(batch[0], float):
         return torch.DoubleTensor(batch)
-    elif isinstance(batch[0], string_classes):
+    elif isinstance(batch[0], str):
         return batch
     elif isinstance(batch[0], collections.abc.Mapping):
         return {key: imed_collate([d[key] for d in batch]) for key in batch[0]}

requirements.txt

@@ -11,16 +11,18 @@ imageio~=2.19
 joblib~=1.0
 matplotlib>=3.3.0
 nibabel~=3.2
-onnxruntime~=1.7
+onnx
+onnxruntime>=1.7.0,<1.16.0
 pandas>=1.1,<1.5.0
-pybids>=0.14.0
+Pillow<10.0.0
+pybids>=0.14.0,<0.15.6
 scikit-learn>=0.20.3
 scikit-image~=0.17
 seaborn~=0.11
 tensorboard>=1.15.0
 tqdm>=4.30
 scipy
 torchio>=0.18.68
-torch>=1.8.1,<=1.11.0
-torchvision>=0.9.1,<=0.12.0
+torch==2.0.0
+torchvision>=0.9.1
 wandb>=0.12.11

testing/unit_tests/test_tensorboard_save.py

@@ -28,9 +28,12 @@ def test_tensorboard_save():
     summary_iterators = [EventAccumulator(str(dname)).Reload() for dname in dpath.iterdir()]
     for i in range(len(summary_iterators)):
         if summary_iterators[i].Tags()['images'] == ['Training/Input', 'Training/Predictions', 'Training/Ground Truth']:
-            input_retrieve = np.array(Image.open(io.BytesIO(summary_iterators[i].Images('Training/Input')[0][2])))
-            pred_retrieve = np.array(Image.open(io.BytesIO(summary_iterators[i].Images('Training/Predictions')[0][2])))
-            gt_retrieve = np.array(Image.open(io.BytesIO(summary_iterators[i].Images('Training/Ground Truth')[0][2])))
+            input_retrieve = np.array(Image.open(io.BytesIO(
+                summary_iterators[i].Images('Training/Input')[0].encoded_image_string)))
+            pred_retrieve = np.array(Image.open(io.BytesIO(
+                summary_iterators[i].Images('Training/Predictions')[0].encoded_image_string)))
+            gt_retrieve = np.array(Image.open(io.BytesIO(
+                summary_iterators[i].Images('Training/Ground Truth')[0].encoded_image_string)))
 
     assert np.allclose(imed_math.rescale_values_array(input_retrieve[:, :, 0], 0, 1), inp[0, 0, :, :])
     assert np.allclose(imed_math.rescale_values_array(pred_retrieve[:, :, 0], 0, 1), pred[0, 0, :, :])