Update ObjectRemovalAlpha dataset card

README.md

@@ -1,4 +1,24 @@
 ---
+annotations_creators:
+  - found
+language:
+  - en
+language_creators:
+  - found
+license: apache-2.0
+pretty_name: ObjectRemovalAlpha
+size_categories:
+  - n<1K
+source_datasets:
+  - lrzjason/ObjectRemovalAlpha
+tags:
+  - object-removal
+  - image-inpainting
+  - image-editing
+  - text-to-image
+task_categories:
+  - image-to-image
+task_ids: []
 dataset_info:
   features:
   - name: text
@@ -22,4 +42,70 @@ configs:
     path: data/train-*
 ---
 
-Reformat from https://huggingface.co/datasets/lrzjason/ObjectRemovalAlpha
+# Dataset Card for ObjectRemovalAlpha
+
+## Dataset Description
+
+- **Homepage:** https://huggingface.co/datasets/lrzjason/ObjectRemovalAlpha
+- **Repository:** https://github.com/creative-graphic-design/huggingface-datasets/tree/main/datasets/ObjectRemovalAlpha
+- **Original Dataset:** https://huggingface.co/datasets/lrzjason/ObjectRemovalAlpha
+- **Hugging Face Dataset:** https://huggingface.co/datasets/creative-graphic-design/ObjectRemovalAlpha
+- **Paper:** Not available in the original dataset release.
+
+### Dataset Summary
+
+ObjectRemovalAlpha is a small object-removal training dataset reformatted from `lrzjason/ObjectRemovalAlpha`. Each example contains a shared text prompt, a ground-truth image, a factual/source image, and a mask image.
+
+The loader groups the original `_G`, `_F`, and `_M` files by numeric ID and exposes them as one row per object-removal example.
+
+### Supported Tasks and Leaderboards
+
+This dataset can be used for image editing, object removal, image inpainting, and mask-conditioned image generation experiments. No public leaderboard is bundled with the source dataset.
+
+### Languages
+
+The text prompts are in English.
+
+## Dataset Structure
+
+### Data Instances
+
+```json
+{
+  "text": "object-removal prompt",
+  "ground_truth_image": "<image>",
+  "factual_image": "<image>",
+  "mask_image": "<image>"
+}
+```
+
+### Data Fields
+
+- `text`: Prompt text shared by the original `_G`, `_F`, and `_M` files.
+- `ground_truth_image`: Target image after removal.
+- `factual_image`: Source image before removal.
+- `mask_image`: Mask image indicating the removal region.
+
+### Data Splits
+
+| Split | Examples |
+| --- | ---: |
+| train | 20 |
+
+## Dataset Creation
+
+### Source Data
+
+The data is loaded from the upstream Hugging Face dataset `lrzjason/ObjectRemovalAlpha` using `snapshot_download`.
+
+### Licensing Information
+
+The upstream dataset is marked Apache-2.0 on Hugging Face Hub.
+
+### Citation Information
+
+No dataset-specific paper or BibTeX citation was found in the upstream dataset release.
+
+### Contributions
+
+This Hugging Face dataset implementation was created by the creative-graphic-design organization to expose `lrzjason/ObjectRemovalAlpha` through a typed dataset loader.