Improve dataset card: add paper link and update metadata

README.md

@@ -1,41 +1,43 @@
 ---
+license: cc-by-4.0
+size_categories:
+- 10M<n<100M
+task_categories:
+- zero-shot-image-classification
+- image-to-text
+- text-to-image
 dataset_info:
   features:
-    - name: jpg
-      dtype: image
-    - name: txt
-      dtype: string
-    - name: njson
-      dtype: string
-    - name: samlens.npy
-      dtype: binary
-    - name: samcat.npy
-      dtype: binary
+  - name: jpg
+    dtype: image
+  - name: txt
+    dtype: string
+  - name: njson
+    dtype: string
+  - name: samlens.npy
+    dtype: binary
+  - name: samcat.npy
+    dtype: binary
   splits:
-    - name: train
-      num_examples: 10968539
+  - name: train
+    num_examples: 10968539
 configs:
-  - config_name: default
-    data_files:
-      - split: train
-        path: "cc12m-train-*.tar"
-license: cc-by-4.0
-task_categories:
-  - zero-shot-image-classification
-  - image-to-text
-  - text-to-image
+- config_name: default
+  data_files:
+  - split: train
+    path: cc12m-train-*.tar
 tags:
-  - clip
-  - webdataset
-  - sam
-  - region-phrase-alignment
-size_categories:
-  - 10M<n<100M
+- clip
+- webdataset
+- sam
+- region-phrase-alignment
 ---
 
 # CC12M with SAM Regions and Parse-Tree Phrases
 
-Pre-processed [CC12M](https://github.com/google-research-datasets/conceptual-12m) dataset for training [PowerCLIP](https://github.com/KMasaki/PowerCLIP).
+This repository contains the pre-processed [CC12M](https://github.com/google-research-datasets/conceptual-12m) dataset used for training **PowerCLIP**, as presented in the paper [PowerCLIP: Powerset Alignment for Contrastive Pre-Training](https://arxiv.org/abs/2511.23170).
+
+The official code is available at [KMasaki0210/PowerCLIP](https://github.com/KMasaki0210/PowerCLIP).
 
 Each sample contains the original image and caption plus two precomputed annotations:
 
@@ -44,7 +46,7 @@ Each sample contains the original image and caption plus two precomputed annotat
 
 ## Format
 
-WebDataset tar archives (2176 shards). Each sample contains:
+The data is stored in WebDataset tar archives (2176 shards). Each sample contains:
 
 ```
 {key}.jpg           # Image
@@ -57,6 +59,8 @@ WebDataset tar archives (2176 shards). Each sample contains:
 
 ## Usage
 
+You can load the dataset using the `webdataset` library:
+
 ```python
 import webdataset as wds
 
@@ -68,7 +72,7 @@ for sample in dataset:
     # by PowerCLIP's data pipeline
 ```
 
-Or use with PowerCLIP directly:
+Or use it with the PowerCLIP training script directly:
 ```bash
 torchrun --nproc_per_node 8 -m training.main \
     --train-data "cc12m-train-{0000..2175}.tar" \
@@ -77,6 +81,6 @@ torchrun --nproc_per_node 8 -m training.main \
 
 ## Source
 
-- Images & captions: [Conceptual 12M](https://github.com/google-research-datasets/conceptual-12m) (CC-BY-4.0)
-- SAM regions: [Segment Anything (ViT-H)](https://github.com/facebookresearch/segment-anything)
-- Parse-tree phrases: [spaCy](https://spacy.io/) `en_core_web_sm`
+- **Images & captions:** [Conceptual 12M](https://github.com/google-research-datasets/conceptual-12m) (CC-BY-4.0)
+- **SAM regions:** [Segment Anything (ViT-H)](https://github.com/facebookresearch/segment-anything)
+- **Parse-tree phrases:** [spaCy](https://spacy.io/) `en_core_web_sm`