Add COCO dataset loading script for proper HuggingFace viewer support

dataset_infos.json

@@ -0,0 +1,44 @@
+{
+    "default": {
+        "description": "USD Side Detection Dataset - COCO format object detection annotations for US Dollar currency with Front/Back classification",
+        "citation": "",
+        "homepage": "https://huggingface.co/datasets/ebowwa/usd-side-coco-annotations",
+        "license": "MIT",
+        "features": {
+            "image_id": {"dtype": "int32", "_type": "Value"},
+            "image": {"_type": "Image"},
+            "width": {"dtype": "int32", "_type": "Value"},
+            "height": {"dtype": "int32", "_type": "Value"},
+            "objects": {
+                "feature": {
+                    "id": {"dtype": "int64", "_type": "Value"},
+                    "area": {"dtype": "float64", "_type": "Value"},
+                    "bbox": {"feature": {"dtype": "float64", "_type": "Value"}, "_type": "Sequence"},
+                    "category": {"names": [
+                        "USD-Dollar-cash-counting--c-TiEZ",
+                        "100USD-Back", "100USD-Front",
+                        "10USD", "10USD-Back", "10USD-Front",
+                        "1USD", "1USD-Back", "1USD-Front",
+                        "20USD-Back", "20USD-Front",
+                        "50USD-Back", "50USD-Front",
+                        "5USD-Back", "5USD-Front",
+                        "Counterfeit 100 USD Back", "Counterfeit 100 USD Front",
+                        "Counterfeit 10USD", "Counterfeit 10USD Front", "Counterfeit 10USD Back",
+                        "Counterfeit 1USD Front", "Counterfeit 1USD Back",
+                        "Counterfeit 20USD Front", "Counterfeit 20USD Back",
+                        "Counterfeit 50USD", "Counterfeit 50USD Front", "Counterfeit 50USD Back",
+                        "Counterfeit 5USD", "Counterfeit 5USD Front", "Counterfeit 5USD Back"
+                    ], "_type": "ClassLabel"}
+                },
+                "_type": "Sequence"
+            }
+        },
+        "splits": {
+            "train": {"name": "train", "num_bytes": 2800000000, "num_examples": 2671},
+            "validation": {"name": "validation", "num_bytes": 600000000, "num_examples": 597},
+            "test": {"name": "test", "num_bytes": 350000000, "num_examples": 350}
+        },
+        "download_size": 3750000000,
+        "dataset_size": 3750000000
+    }
+}

usd_side_coco_annotations.py

@@ -0,0 +1,129 @@
+"""USD Side Detection Dataset - COCO format object detection."""
+
+import json
+import os
+from pathlib import Path
+
+import datasets
+from datasets import DatasetBuilder, DownloadManager, SplitGenerator
+from datasets.features import ClassLabel, Features, Image, Sequence, Value
+
+
+_DESCRIPTION = """
+USD Side Detection Dataset with Front/Back classification.
+
+A COCO-format dataset for detecting US Dollar currency and classifying 
+whether the front or back side is visible.
+
+- 3,618 images (all with annotations)
+- 3,746 annotations  
+- 24 classes (12 regular + 12 counterfeit USD)
+- 100% Front/Back classified
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/ebowwa/usd-side-coco-annotations"
+
+_LICENSE = "MIT"
+
+_CATEGORIES = [
+    "100USD-Back", "100USD-Front",
+    "10USD-Back", "10USD-Front", 
+    "1USD-Back", "1USD-Front",
+    "20USD-Back", "20USD-Front",
+    "50USD-Back", "50USD-Front",
+    "5USD-Back", "5USD-Front",
+    "Counterfeit 100 USD Back", "Counterfeit 100 USD Front",
+    "Counterfeit 10USD Back", "Counterfeit 10USD Front",
+    "Counterfeit 1USD Back", "Counterfeit 1USD Front",
+    "Counterfeit 20USD Back", "Counterfeit 20USD Front",
+    "Counterfeit 50USD Back", "Counterfeit 50USD Front",
+    "Counterfeit 5USD Back", "Counterfeit 5USD Front",
+]
+
+
+class UsdSideCocoAnnotations(DatasetBuilder):
+    """USD Side Detection Dataset."""
+
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=Features({
+                "image_id": Value("int32"),
+                "image": Image(),
+                "width": Value("int32"),
+                "height": Value("int32"),
+                "objects": Sequence({
+                    "id": Value("int64"),
+                    "area": Value("float64"),
+                    "bbox": Sequence(Value("float64"), length=4),
+                    "category": ClassLabel(names=_CATEGORIES),
+                }),
+            }),
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+        )
+
+    def _split_generators(self, dl_manager: DownloadManager):
+        data_dir = dl_manager.download_and_extract(
+            "https://huggingface.co/datasets/ebowwa/usd-side-coco-annotations/resolve/main/"
+        ) if dl_manager else Path(".")
+        
+        return [
+            SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={"split_dir": os.path.join(data_dir, "train")},
+            ),
+            SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                gen_kwargs={"split_dir": os.path.join(data_dir, "valid")},
+            ),
+            SplitGenerator(
+                name=datasets.Split.TEST,
+                gen_kwargs={"split_dir": os.path.join(data_dir, "test")},
+            ),
+        ]
+
+    def _generate_examples(self, split_dir):
+        ann_file = os.path.join(split_dir, "_annotations.coco.json")
+        
+        with open(ann_file) as f:
+            coco = json.load(f)
+        
+        # Build category mapping
+        cat_id_to_name = {c["id"]: c["name"] for c in coco["categories"]}
+        cat_name_to_idx = {name: idx for idx, name in enumerate(_CATEGORIES)}
+        
+        # Group annotations by image_id
+        img_to_anns = {}
+        for ann in coco["annotations"]:
+            img_id = ann["image_id"]
+            if img_id not in img_to_anns:
+                img_to_anns[img_id] = []
+            img_to_anns[img_id].append(ann)
+        
+        # Generate examples
+        for img in coco["images"]:
+            img_id = img["id"]
+            img_path = os.path.join(split_dir, img["file_name"])
+            
+            anns = img_to_anns.get(img_id, [])
+            objects = []
+            for ann in anns:
+                cat_name = cat_id_to_name.get(ann["category_id"])
+                if cat_name in cat_name_to_idx:
+                    objects.append({
+                        "id": ann["id"],
+                        "area": ann["area"],
+                        "bbox": ann["bbox"],
+                        "category": cat_name_to_idx[cat_name],
+                    })
+            
+            yield img_id, {
+                "image_id": img_id,
+                "image": img_path,
+                "width": img["width"],
+                "height": img["height"],
+                "objects": objects,
+            }