Merge pull request #2 from pj-ms/pinjin/index-from-zero

make index for img, class or annotation can start from zero

README.md

@@ -75,6 +75,8 @@ Here is an example with explanation of what a `DatasetInfo` looks like for coco
 
 ##### Coco JSON - Image classification
 
+Here is one example of the train.json, val.json, or test.json in the `DatasetInfo` above. Note that the `"id"` for `images`, `annotations` and `categories` should be consecutive integers, **starting from 1**. Note that our lib might work with id starting from 0, but many tools like [CVAT](https://github.com/openvinotoolkit/cvat/issues/2085) and official [COCOAPI](https://github.com/cocodataset/cocoapi/issues/507) will fail.
+
 ``` {json}
 {
   "images": [{"id": 1, "width": 224.0, "height": 224.0, "file_name": "train_images.zip@siberian-kitten.jpg"},
@@ -152,6 +154,8 @@ Each rows in the index file (`index_path`) is:
 <image_filepath> <comma-separated-label-indices>
 ```
 
+Note that the class/label index should start from zero.
+
 Example:
 
 ``` {txt}
@@ -168,6 +172,8 @@ The index file for OD is slightly different from IC. Each rows in the index file
 <image_filepath> <label_filepath>
 ```
 
+Same with classification, the class/label index should start from 0.
+
 Example for `train_images.txt`:
 
 ``` {txt}

setup.py

@@ -1,7 +1,7 @@
 import setuptools
 from os import path
 
-VERSION = '0.1.0'
+VERSION = '0.1.1'
 
 # Get the long description from the README file
 here = path.abspath(path.dirname(__file__))

tests/test_create_dataset_manifest.py

@@ -211,6 +211,31 @@ class TestCreateCocoDatasetManifest(unittest.TestCase):
             self.assertEqual(dataset_manifest.images[0].labels, [0])
             self.assertEqual(dataset_manifest.images[1].labels, [0, 1])
 
+    def test_index_can_start_from_zero(self):
+        manifest_dict = {
+            "images": [{"id": 0, "width": 224.0, "height": 224.0, "file_name": "siberian-kitten.jpg"},
+                       {"id": 1, "width": 224.0, "height": 224.0, "file_name": "kitten 3.jpg"}],
+            "annotations": [
+                {"id": 0, "category_id": 0, "image_id": 0},
+                {"id": 1, "category_id": 0, "image_id": 1},
+                {"id": 2, "category_id": 1, "image_id": 1}
+            ], "categories": [{"id": 0, "name": "cat"}, {"id": 1, "name": "dog"}]
+        }
+        dataset_dict = copy.deepcopy(self.DATASET_INFO_DICT)
+        with tempfile.TemporaryDirectory() as tempdir:
+            dataset_dict['root_folder'] = ''
+            dataset_dict['type'] = 'classification_multilabel'
+            coco_file_path = os.path.join(tempdir, 'test.json')
+            with open(coco_file_path, 'w') as f:
+                json.dump(manifest_dict, f)
+
+            dataset_manifest = CocoManifestAdaptor.create_dataset_manifest(coco_file_path, DatasetTypes.IC_MULTILABEL)
+            self.assertIsInstance(dataset_manifest, DatasetManifest)
+            self.assertEqual(len(dataset_manifest.images), 2)
+            self.assertEqual(len(dataset_manifest.labelmap), 2)
+            self.assertEqual(dataset_manifest.images[0].labels, [0])
+            self.assertEqual(dataset_manifest.images[1].labels, [0, 1])
+
     def test_object_detection(self):
         manifest_dict = {
             "images": [{"id": 1, "width": 224.0, "height": 224.0, "file_name": "siberian-kitten.jpg"},

vision_datasets/common/data_manifest.py

@@ -439,8 +439,12 @@ class CocoManifestAdaptor:
 
         images_by_id = {img['id']: ImageDataManifest(img['id'], get_full_sas_or_path(img['file_name']), img['width'], img['height'], []) for img in coco_manifest['images']}
 
+        label_dict_by_id = {cate['id']: cate['name'] for cate in coco_manifest['categories']}
+        label_starting_idx = min(label_dict_by_id.keys())
+        labelmap = [label_dict_by_id[i + label_starting_idx] for i in range(len(label_dict_by_id))]
+
         for annotation in coco_manifest['annotations']:
-            c_id = annotation['category_id'] - 1
+            c_id = annotation['category_id'] - label_starting_idx
             if 'bbox' in annotation:
                 label = [c_id] + annotation['bbox']
             else:
@@ -449,7 +453,4 @@ class CocoManifestAdaptor:
         images = [x for x in images_by_id.values()]
         images.sort(key=lambda x: x.id)
 
-        label_dict_by_id = {cate['id']: cate['name'] for cate in coco_manifest['categories']}
-        labelmap = [label_dict_by_id[i + 1] for i in range(len(label_dict_by_id))]
-
         return DatasetManifest(images, labelmap, data_type)