updated readme

README.md

@@ -1,4 +1,85 @@
----
+
+
+# Generative-VQA-V2-Curated
+
+A curated, balanced, and cleaned version of the VQA v2 dataset specifically optimized for **Generative Visual Question Answering**.
+
+This dataset transforms the standard VQA task into a generative challenge by removing "yes/no" shortcuts and balancing answer distributions to prevent model over-fitting on dominant classes.
+
+## Dataset Summary
+
+The primary goal of this curated set is to provide a "clean" signal for training multimodal models by:
+
+1. **Eliminating Binary Biases:** Removed all "yes/no" and "unknown" style answers.
+2. **Balancing Classes:** Capped samples at **600 per answer** to ensure the model learns a diverse vocabulary.
+3. **Filtering Ambiguity:** Removed generic questions (e.g., "What is this?") to focus on specific visual grounding.
+
+## Dataset Statistics
+
+* **Total QA Pairs:** 135,268
+* **Unique Answer Classes:** 1,251
+* **Source Images:** COCO Train 2014
+* **Minimum Frequency per Answer:** 20
+* **Maximum Samples per Answer:** 600
+
+## 🛠 Curation Logic
+
+The dataset was generated using the following filtering pipeline:
+
+* **Consensus-Based:** Only the majority-vote answer from the 10 human annotators is used.
+* **Exclusion List:** * Boolean: `yes`, `no`
+* Uncertainty: `unknown`, `none`, `n/a`, `cant tell`, `not sure`
+
+
+* **Ambiguity Filter:** Removed questions containing "what is in the image", "what is this", "what is that", or "what do you see".
+* **Conciseness:** Answers are restricted to **5 words** and **30 characters** or fewer.
+
+## Structure
+
+```text
+.
+├── images/               # Curated COCO images (JPG)
+├── qa_pairs.json         # Full metadata (JSON)
+├── metadata.csv          # Metadata (CSV) for easy loading
+└── README.md
+
+```
+
+### Metadata Fields
+
+* `image_id`: Original COCO Image ID.
+* `question_id`: Original VQA v2 Question ID.
+* `question`: The natural language question.
+* `answer`: The curated ground-truth answer.
+* `image_path`: Path relative to the dataset root.
+
+##  Usage
+
+### Loading with Hugging Face `datasets`
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("Deva8/Generative-VQA-V2-Curated")
+
+```
+
+### Quick Look (Pandas)
+
+```python
+import pandas as pd
+df = pd.read_csv("metadata.csv")
+print(df['answer'].value_counts().head(10)) # Check balancing
+
+```
+
+## License & Attribution
+
+This dataset is a derivative work of the **VQA v2 Dataset** and the **COCO Dataset**.
+
+* **Images:** [COCO Consortium (CC BY 4.0)](https://www.google.com/search?q=https://cocodataset.org/%23termsofuse)
+* **Annotations:** [VQA v2 (CC BY 4.0)](https://visualqa.org/download.html)
+
 license: mit
 task_categories:
 - visual-question-answering
@@ -7,5 +88,4 @@ tags:
 - code
 pretty_name: Vqa-Dataset for generative tasks
 size_categories:
-- 10K<n<100K
----
+- 10K<n<100K