Upload README.md with huggingface_hub

README.md

@@ -1,19 +1,40 @@
 ---
-dataset_info:
-  features:
-  - name: image
-    dtype: image
-  - name: mask
-    dtype: image
-  splits:
-  - name: test
-    num_bytes: 24103661
-    num_examples: 280
-  download_size: 24125938
-  dataset_size: 24103661
-configs:
-- config_name: default
-  data_files:
-  - split: test
-    path: data/test-*
+license: cc-by-nc-4.0
+pretty_name: COIFT
+task_categories:
+- image-segmentation
+tags:
+- interactive-segmentation
+- thin-object-segmentation
+- foreground-segmentation
+size_categories:
+- n<1K
 ---
+
+# COIFT
+
+COIFT (COco Instances For Thin objects) is an interactive/thin-object
+segmentation benchmark consisting of 280 images with high-quality binary
+foreground masks. It is used to evaluate segmentation of objects with thin
+structures.
+
+## Dataset structure
+
+- Split: `test` (280 examples) — COIFT is a single benchmark set with no train/test split.
+- Columns:
+  - `image`: the RGB input image (`datasets.Image`).
+  - `mask`: the binary ground-truth foreground mask, single-channel (`datasets.Image`).
+
+Images and masks are aligned 1:1 by filename stem.
+
+## Source & credit
+
+Redistributed from the **thin-object-selection** repository accompanying the
+paper *"Deep Interactive Thin Object Selection"* (Liew et al.).
+
+- Repository: https://github.com/liewjunhao/thin-object-selection
+
+## License
+
+Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC-4.0),
+following the source repository. Non-commercial use only.