convert 16 bit greyscale images to rgb

scripts/dvc.lock

@@ -7,20 +7,20 @@ stages:
     outs:
     - path: ../vectors
       hash: md5
-      md5: 381c0f3d25a23daa7a3b741e9bb41a1d.dir
+      md5: 6be45d2101d76bd5f29c6ec858aecc8a.dir
       size: 13566692
       nfiles: 5
   cluster:
     cmd: python cluster.py
     deps:
     - path: ../vectors
       hash: md5
-      md5: b56c902defd862c4326adc981abea25c.dir
+      md5: 18fe90eccd3a31d63eb81a0db701e375.dir
       size: 13566692
       nfiles: 5
     outs:
     - path: ../models
       hash: md5
-      md5: d751713988987e9331980363e24189ce.dir
-      size: 0
-      nfiles: 0
+      md5: 17573502996bd38fde3c4304cb3d1d3d.dir
+      size: 84672
+      nfiles: 1

scripts/dvc.yaml

@@ -11,8 +11,3 @@ stages:
     - ../vectors
     outs:
     - ../models
-
-artifacts:
-  kmeans_cluster: # artifact ID (name)
-    path: ../models/kmeans.pkl
-    type: model

scripts/image_embeddings.py

@@ -1,4 +1,5 @@
-"""Try to use the scivision pretrained model and tools against this collection"""
+"""Extract and store image embeddings from a collection in s3,
+using an off-the-shelf pre-trained model"""
 
 import os
 import logging

scripts/image_metadata.py

@@ -1,11 +1,4 @@
-"""Heavy-handed approach to create image metadata in usable with `intake`,
-for trial use with `scivision`:
-https://scivision.readthedocs.io/en/latest/api.html#scivision.io.reader.load_dataset
-https://intake.readthedocs.io/en/latest/catalog.html#yaml-format
-
-See also https://github.com/intake/intake-stac
-Via https://gallery.pangeo.io/repos/pangeo-data/pangeo-tutorial-gallery/intake.html#Build-an-intake-catalog
-
+"""Create a basic index for the images in an s3 collection
 """
 
 import yaml

scripts/params.yaml

@@ -1,4 +1,4 @@
 cluster:
   n_clusters: 5
 
-collection: untagged-images-lana
+collection: untagged-images-wala

src/cyto_ml/data/image.py

@@ -27,6 +27,14 @@ def prepare_image(image: Image) -> torch.Tensor:
     a) Converts the image data to a PyTorch tensor
     b) Accepts a single image or batch (no need for torch.stack)
     """
+    # Flow Cytometer images are 16-bit greyscale
+    # https://stackoverflow.com/questions/18522295/python-pil-change-greyscale-tif-to-rgb
+    # TODO revisit
+
+    if image.mode == "I;16":
+        image.point(lambda p: p * 0.0039063096, mode="RGB")
+        image = image.convert("RGB")
+
     tensor_image = transforms.ToTensor()(image)
 
     # Single image, add a batch dimension

tests/conftest.py

@@ -18,6 +18,12 @@ def image_dir(fixture_dir):
     return os.path.join(fixture_dir, "test_images")
 
 
+@pytest.fixture
+def greyscale_image(fixture_dir):
+    """Directory with single plankton images"""
+    return os.path.join(fixture_dir, "greyscale", "TC18_280524_4840.tif")
+
+
 @pytest.fixture
 def single_image(image_dir):
     # The file naming conventions were like this when i got here

tests/test_prepare_image.py

@@ -17,5 +17,12 @@ def test_single_image(single_image):
 
     assert torch.all((prepared_image >= 0.0) & (prepared_image <= 1.0))
 
+def test_greyscale_image(greyscale_image):
+
+    # Tensorise the image (potentially normalise if we have useful values)
+    prepared_image = load_image(greyscale_image)
+    assert torch.all((prepared_image >= 0.0) & (prepared_image <= 1.0))
+
+
 if __name__ == "__main__":
     pytest.main()