Upload MOTIF dataset

.gitattributes

@@ -1,2 +1,3 @@
 *.png filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -19,7 +19,21 @@ size_categories:
 - 1K<n<10K
 configs:
 - config_name: default
-  drop_labels: true
+  data_files:
+  - split: train
+    path: data/train.parquet
+dataset_info:
+  features:
+  - name: image
+    dtype: image
+  - name: id
+    dtype: string
+  - name: context
+    dtype: string
+  - name: focus
+    dtype: string
+  - name: image_id
+    dtype: string
 ---
 
 # MOTIF
@@ -30,21 +44,18 @@ The uploaded archive contains 1,125 examples from the L2Corpus release. Each exa
 
 ## Dataset Structure
 
-This Hugging Face repository is organized as an image dataset:
+This Hugging Face repository is organized as a Parquet image dataset:
 
 ```text
-train/
-  metadata.csv
-  images/
-    *.png
+data/
+  train.parquet
 ```
 
-The `metadata.csv` file contains the original text fields plus the image path used by the Hugging Face image dataset loader.
+The Parquet file embeds each annotated PNG image in an `image` column and keeps the original text metadata alongside it.
 
 ## Fields
 
-- `image`: image loaded from `file_name` by Hugging Face.
-- `file_name`: relative path to the annotated image.
+- `image`: annotated image associated with the focus word and context.
 - `id`: original example identifier.
 - `context`: reading context sentence.
 - `focus`: complex word or target focus word in the context.

train/images/fw90_10038_495852_annotated.png → data/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c36abdc1b379a3d6e92bc6a67e851c3e3bf04662be289cf6c6f8a7c2af388bba
-size 201102
+oid sha256:7ee5ee51a84b0d521ad6e07ce43f916c127e733de25dda0bc0c455a0d09caf56
+size 617069569