Fix ccagt to coco ps masks sizes (#52)

CCAgT_utils/converters/CCAgT.py

@@ -474,17 +474,16 @@ def to_PS_COCO(self, categories_infos: CategoriesInfos, out_dir: str, split_by_s
 
         print(
             f'Start compute generate panoptic annotations and masks for {len(images_ids)} files using {cpu_num} cores with '
-            '{len(images_ids_splitted[0])} files per core...',
+            f'{len(images_ids_splitted[0])} files per core...',
         )
 
         workers = multiprocessing.Pool(processes=cpu_num)
         processes = []
-        output_template = np.zeros((self.IMAGE_HEIGHT, self.IMAGE_WIDTH, 3), dtype=np.uint8)
         for images_ids in images_ids_splitted:
             if len(images_ids) == 0:
                 continue
             df_to_process = self.df.loc[self.df['image_id'].isin(images_ids), :]
-            p = workers.apply_async(single_core_to_PS_COCO, (df_to_process, out_dir, output_template, split_by_slide))
+            p = workers.apply_async(single_core_to_PS_COCO, (df_to_process, out_dir, split_by_slide))
             processes.append(p)
 
         annotations_panoptic = []
@@ -501,12 +500,19 @@ def copy(self) -> CCAgT:
 def single_core_to_PS_COCO(
     df: pd.DataFrame,
     out_dir: str,
-    output_template: np.ndarray,
     split_by_slide: bool,
 ) -> list[dict[str, Any]]:
     annotations_panoptic = []
 
     _out_dir = out_dir
+
+    if df['image_width'].nunique() == df['image_height'].nunique() == 1:
+        w = int(df['image_width'].unique()[0])
+        h = int(df['image_height'].unique()[0])
+        output_template = np.zeros((h, w, 3), dtype=np.uint8)
+    else:
+        output_template = None
+
     for img_id, df_by_img in df.groupby('image_id'):
         img_name = df_by_img.iloc[0]['image_name']
         output_basename = basename(img_name) + '.png'
@@ -525,7 +531,14 @@ def single_core_to_PS_COCO(
         ])
 
         segments_info = []
-        out = output_template.copy()
+
+        if isinstance(output_template, np.ndarray):
+            out = output_template.copy()
+        else:
+            w = int(df_by_img['image_width'].unique()[0])
+            h = int(df_by_img['image_height'].unique()[0])
+            out = np.zeros((h, w, 3), dtype=np.uint8)
+
         for ann in annotations_sorted:
             out = draw_annotation(out, ann, ann.color.rgb, out.shape[:2])
             segments_info.append({

setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = CCAgT_utils
-version = 0.2.2-alpha
+version = 0.2.3-alpha
 description = A framework of utilities to help at the use of the CCAgT dataset
 long_description = file: README.md
 long_description_content_type = text/markdown

tests/conftest.py

@@ -281,8 +281,11 @@ def ccagt_df_single_nucleus(nucleus_ex):
 
 
 @pytest.fixture
-def ccagt_ann_single_nucleus(nucleus_ex):
+def ccagt_ann_single_nucleus(nucleus_ex, shape):
     d = pd.DataFrame([create.row_CCAgT(nucleus_ex, 1, 'C_xx1')])
+    d['image_width'] = shape[1]
+    d['image_height'] = shape[0]
+
     return CCAgT.CCAgT(d)
 
 

tests/converters/CCAgT_test.py

@@ -275,15 +275,14 @@ def test_generate_masks(ccagt_ann_single_nucleus):
         assert os.path.isfile(os.path.join(tmp_dir, 'C/C_xx1' + '.png'))
 
 
-def test_single_core_to_PS_COCO(ccagt_ann_single_nucleus, tmpdir, shape):
+def test_single_core_to_PS_COCO(ccagt_ann_single_nucleus, tmpdir):
     ccagt_ann_single_nucleus.df['image_name'] = 'C_xx1'
     ccagt_ann_single_nucleus.df['area'] = ccagt_ann_single_nucleus.geometries_area()
     ccagt_ann_single_nucleus.df['image_id'] = ccagt_ann_single_nucleus.generate_ids(ccagt_ann_single_nucleus.df['image_name'])
     ccagt_ann_single_nucleus.df['iscrowd'] = 0
     ccagt_ann_single_nucleus.df['color'] = Color(21, 62, 125)
 
-    template = np.zeros((shape[0], shape[1], 3), dtype=np.uint8)
-    out = CCAgT.single_core_to_PS_COCO(ccagt_ann_single_nucleus.df, tmpdir, template, False)
+    out = CCAgT.single_core_to_PS_COCO(ccagt_ann_single_nucleus.df, tmpdir, False)
 
     check1 = all(y in out[0] for y in {'image_id', 'file_name', 'segments_info'})
     assert check1
@@ -294,10 +293,28 @@ def test_single_core_to_PS_COCO(ccagt_ann_single_nucleus, tmpdir, shape):
 
     ccagt_ann_single_nucleus.df['slide_id'] = ccagt_ann_single_nucleus.get_slide_id()
     tmpdir.mkdir('C')
-    CCAgT.single_core_to_PS_COCO(ccagt_ann_single_nucleus.df, tmpdir, template, True)
+    CCAgT.single_core_to_PS_COCO(ccagt_ann_single_nucleus.df, tmpdir, True)
     assert len(os.listdir(os.path.join(tmpdir, 'C/'))) > 0
 
 
+def test_single_core_to_PS_COCO_multisizes(ccagt_ann_multi, tmpdir):
+    ccagt_ann_multi.df['area'] = ccagt_ann_multi.geometries_area()
+    ccagt_ann_multi.df['image_id'] = ccagt_ann_multi.generate_ids(ccagt_ann_multi.df['image_name'])
+    ccagt_ann_multi.df['iscrowd'] = 0
+    ccagt_ann_multi.df['color'] = Color(21, 62, 125)
+
+    ccagt_ann_multi.df['image_width'] = 1000
+    ccagt_ann_multi.df['image_height'] = 1000
+
+    ccagt_ann_multi.df.loc[ccagt_ann_multi.df['image_id'] == 1, 'image_height'] = 2000
+
+    ccagt_ann_multi.df['geo_type'] = ccagt_ann_multi.df['geometry'].apply(lambda g: g.geom_type).tolist()
+    ccagt_ann_multi.df = ccagt_ann_multi.df[ccagt_ann_multi.df['geo_type'] != 'Point']
+    CCAgT.single_core_to_PS_COCO(ccagt_ann_multi.df, tmpdir, False)
+
+    assert len(os.listdir(tmpdir)) > 0
+
+
 def test_CCAgT_to_PS_COCO(ccagt_ann_single_nucleus, categories_infos, tmpdir):
     ccagt_ann_single_nucleus.df['image_name'] = 'C_xx1'
     ccagt_ann_single_nucleus.df['area'] = ccagt_ann_single_nucleus.geometries_area()