Merge pull request #91 from jumutc/master

Resolving grayscale image handling & conversions

miscnn/data_loading/interfaces/image_io.py

@@ -95,15 +95,18 @@ def load_image(self, index):
         # Load image from file
         img_raw = Image.open(os.path.join(img_path, "imaging" + "." + \
                                           self.img_format))
-        # Convert image to rgb or grayscale
-        if self.img_type == "grayscale":
+        # Convert image to rgb or grayscale if needed
+        if self.img_type == "grayscale" and len(img_raw.getbands()) > 1:
             img_pil = img_raw.convert("LA")
-        elif self.img_type == "rgb":
+        elif self.img_type == "rgb" and img_raw.mode != "RGB":
             img_pil = img_raw.convert("RGB")
+        else:
+            img_pil = img_raw
         # Convert Pillow image to numpy matrix
         img = np.array(img_pil)
-        # Keep only intensity for grayscale images
-        if self.img_type =="grayscale" : img = img[:,:,0]
+        # Keep only intensity for grayscale images if needed
+        if self.img_type == "grayscale" and len(img.shape) > 2:
+            img = img[:, :, 0]
         # Return image
         return img, {"type": "image"}
 

tests/test_iointerfaces.py

@@ -43,18 +43,24 @@ def setUpClass(self):
         # Create image and segmentation
         np.random.seed(1234)
         img = np.random.rand(16, 16, 16) * 255
-        self.img = img.astype(int)
+        self.img = img.astype(np.uint8)
         seg = np.random.rand(16, 16, 16) * 3
-        self.seg = seg.astype(int)
+        self.seg = seg.astype(np.uint8)
         # Initialize temporary directory
         self.tmp_data = tempfile.TemporaryDirectory(prefix="tmp.miscnn.",
                                                     suffix=".data")
         # Write PNG sample with image and segmentation
-        path_sample_image = opj(self.tmp_data.name, "image")
+        path_sample_image = opj(self.tmp_data.name, "image_01")
         os.mkdir(path_sample_image)
-        img_pillow = Image.fromarray(img[:,:,0].astype(np.uint8))
+        img_pillow = Image.fromarray(self.img[:,:,0])
         img_pillow.save(opj(path_sample_image, "imaging.png"))
-        seg_pillow = Image.fromarray(seg[:,:,0].astype(np.uint8))
+        seg_pillow = Image.fromarray(self.seg[:,:,0])
+        seg_pillow.save(opj(path_sample_image, "segmentation.png"))
+        path_sample_image = opj(self.tmp_data.name, "image_02")
+        os.mkdir(path_sample_image)
+        img_pillow = Image.fromarray(self.img[:, :, :3])
+        img_pillow.save(opj(path_sample_image, "imaging.png"))
+        seg_pillow = Image.fromarray(self.seg[:, :, 0])
         seg_pillow.save(opj(path_sample_image, "segmentation.png"))
         # Write NIfTI sample with image and segmentation
         path_sample_nifti = opj(self.tmp_data.name, "nifti")
@@ -76,23 +82,28 @@ def test_IOI_IMAGE_creation(self):
         interface = Image_interface()
     # Initialization
     def test_IOI_IMAGE_initialize(self):
-        interface = Image_interface(pattern="image")
+        interface = Image_interface(pattern="image_\\d+")
         sample_list = interface.initialize(self.tmp_data.name)
-        self.assertEqual(len(sample_list), 1)
-        self.assertEqual(sample_list[0], "image")
+        self.assertEqual(len(sample_list), 2)
+        self.assertTrue("image_01" in sample_list)
+        self.assertTrue("image_02" in sample_list)
     # Loading Images and Segmentations
     def test_IOI_IMAGE_loading(self):
-        interface = Image_interface(pattern="image")
-        sample_list = interface.initialize(self.tmp_data.name)
-        img = interface.load_image(sample_list[0])
-        seg = interface.load_segmentation(sample_list[0])
-        self.assertTrue(np.array_equal(img[0], self.img[:,:,0]))
-        self.assertTrue(np.array_equal(seg, self.seg[:,:,0]))
+        interface = Image_interface(pattern="image_\\d+")
+        sample_list = sorted(interface.initialize(self.tmp_data.name))
+        img_01 = interface.load_image(sample_list[0])
+        seg_01 = interface.load_segmentation(sample_list[0])
+        self.assertTrue(np.array_equal(img_01[0], self.img[:, :, 0]))
+        self.assertTrue(np.array_equal(seg_01, self.seg[:,:,0]))
+        img_02 = interface.load_image(sample_list[1])
+        seg_02 = interface.load_segmentation(sample_list[1])
+        true_img_02 = np.array(Image.fromarray(self.img[:, :, :3]).convert('LA'))
+        self.assertTrue(np.array_equal(img_02[0], true_img_02[:,:,0]))
+        self.assertTrue(np.array_equal(seg_02, self.seg[:, :, 0]))
     # NIFTI_interface - Loading and Storage of Predictions
     def test_IOI_IMAGE_predictionhandling(self):
         interface = Image_interface(pattern="image")
-        sample_list = interface.initialize(self.tmp_data.name)
-        sample = MIScnn_sample.Sample("pred.image", self.img[:,:,0], 1, 2);
+        sample = MIScnn_sample.Sample("pred.image", self.img[:,:,0], 1, 2)
         sample.add_prediction(self.seg[:,:,0]);
         interface.save_prediction(sample, self.tmp_data.name)
         pred = interface.load_prediction("pred.image", self.tmp_data.name)