Upload folder using huggingface_hub

.gitattributes

@@ -1,3 +1,4 @@
 *.png filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -13,6 +13,13 @@ tags:
   - fibroblast
 size_categories:
   - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: example
+        path: data/example.parquet
+      - split: test
+        path: data/test.parquet
 ---
 
 # Micro-OD
@@ -135,10 +142,20 @@ Annotations are stored as [JSON Lines](https://jsonlines.org/) (`.jsonl`) files
 
 ## Usage
 
+Before pushing to HuggingFace Hub, generate the Parquet files from the raw annotations:
+
+```bash
+pip install datasets pillow pyarrow
+python scripts/generate_parquet.py
+# Outputs: data/example.parquet, data/test.parquet
+```
+
+Then load the dataset normally:
+
 ```python
 from datasets import load_dataset
 
-ds = load_dataset("path/to/Micro-OD")
+ds = load_dataset("stumbledparams/Micro-OD")
 
 # Access splits
 example_split = ds["example"]   # 40 images — few-shot support set
@@ -149,12 +166,12 @@ test_split    = ds["test"]       # 212 images — evaluation query set
 #   image_id    — "<subdataset>/images/<filename>"
 #   subdataset  — one of: BBBC, BCCD, LIVECell, NIH-3T3
 #   objects     — dict with keys:
-#                   bbox     : list of [x_min, y_min, width, height]  (COCO format, float)
-#                   label    : list of int  (index into the 10-class label set)
+#                   bbox     : list of [x_min, y_min, width, height]  (COCO format, float32)
+#                   label    : list of int  (ClassLabel index, 0–9)
 #                   category : list of str  (class name)
 ```
 
-> **Note on bbox format:** The loading script converts the raw `[[x_min, y_min], [x_max, y_max]]` annotation format into COCO-style `[x_min, y_min, width, height]` float lists. If you consume the `annotation.jsonl` files directly, refer to the [Annotation Format](#annotation-format) section for the raw coordinate convention.
+> **Note on bbox format:** The Parquet files store bboxes in COCO format `[x_min, y_min, width, height]` as float32. The raw `annotation.jsonl` files use `[[x_min, y_min], [x_max, y_max]]` (top-left / bottom-right pixel coordinates) — see [Annotation Format](#annotation-format).
 
 ## Dataset Statistics
 

data/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:383c5c1aed33a947efd4820b7f76a1dbefc2f84c7f6550a1ba08b0edd6834f74
+size 264405