Upload README.md with huggingface_hub

README.md

@@ -1,8 +1,6 @@
 ---
 language: en
 license: apache-2.0
-size_categories:
-- 10K<n<11K
 task_categories:
 - visual-question-answering
 - object-detection
@@ -11,12 +9,11 @@ tags:
 - drone
 - reasoning
 - vqa
-- surveillance
 - chain-of-thought
-- military
-- object-detection
+- qwen-vl
+- counter-uas
 configs:
-- config_name: default
+- config_name: v2
   data_files:
   - split: train
     path: data/train-*
@@ -24,134 +21,43 @@ configs:
     path: data/validation-*
   - split: test
     path: data/test-*
-dataset_info:
-  features:
-  - name: image
-    dtype: image
-  - name: image_path
-    dtype: string
-  - name: source
-    dtype: string
-  - name: ground_truth
-    dtype: string
-  - name: ground_truth_norm
-    dtype: string
-  - name: messages
-    dtype: string
-  - name: categories
-    dtype: string
-  - name: dup_weight
-    dtype: float64
-  - name: model
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 3404750364
-    num_examples: 13423
-  - name: validation
-    num_bytes: 180451113
-    num_examples: 743
-  - name: test
-    num_bytes: 172807697
-    num_examples: 743
-  download_size: 3535461228
-  dataset_size: 3758009174
 ---
 
-# UAV Drone Reasoning VQA Dataset
-
-A visual question answering dataset for drone/UAV detection and reasoning, featuring
-chain-of-thought reasoning with **thinking → answer** pairs.
-
-## Dataset Description
-
-This dataset contains **10,000** surveillance images annotated with:
-- **72,019** question-answer pairs (avg 7.2 per image)
-- Ground-truth bounding boxes with UAV class labels
-- Chain-of-thought reasoning ("thinking" field) for each answer
-
-### Intended Use
-Training and evaluating multimodal models for:
-- UAV/drone detection and classification in surveillance imagery
-- Visual reasoning with chain-of-thought explanations
-- Scene understanding (weather, visibility, lighting conditions)
-- Threat assessment and detection difficulty analysis
-
-## Dataset Structure
-
-Each row contains:
-| Field | Type | Description |
-|-------|------|-------------|
-| `image` | Image | The surveillance image |
-| `image_path` | string | Original relative path |
-| `ground_truth` | string (JSON) | Bounding boxes and class labels |
-| `conversations` | string (JSON) | List of QA pairs with thinking/reasoning |
-
-### Conversation Format
-```json
-[
-  {
-    "question": "Is there a UAV visible in the sky?",
-    "thinking": "Scanning the sky reveals a small object...",
-    "answer": "Yes, a UAV is present."
-  }
-]
-```
-
-### Question Categories
-- **Detection**: Presence/absence of UAVs
-- **Counting**: Number of objects
-- **Classification**: UAV type/model identification
-- **Localization**: Bounding box coordinates
-- **Scene Analysis**: Weather, lighting, visibility
-- **Characteristics**: Visual features of detected UAVs
-- **Risk Assessment**: Threat level and detection difficulty
-
-## UAV Classes
-
-| Class | Count |
-|-------|-------|
-| mq9_reaper | 3039 |
-| mohajer_6 | 3016 |
-| shahed_238 | 2995 |
-| shahed_136 | 2962 |
-| dji_mavic | 2954 |
-
-## Source Image Splits
-
-| Split | Images |
-|-------|--------|
-| train | 7003 |
-| val | 2001 |
-| test | 996 |
-
-## Generation Pipeline
-
-QA pairs were generated using **Qwen3.7-Plus** (multimodal) via OpenRouter API,
-with ground-truth YOLO annotations as guidance. Each response was validated for:
-- Valid JSON structure
-- Minimum 3 QA pairs per image
-- Required fields (question, thinking, answer)
+# UAV Drone Reasoning VQA Dataset v2
+
+Multimodal VQA dataset for counter-UAS training with rich chain-of-thought reasoning.
+
+## v2 Improvements
+- Negative and hard-negative samples (birds/aircraft distractors)
+- Category-labeled QA pairs
+- Richer multi-sentence thinking traces
+- Qwen-VL chat `messages` format with `` blocks
+- Stratified train/validation/test splits
+- Duplicate-question downweighting via `dup_weight`
+
+## Stats
+- Records: 14,909
+- QA pairs: 98,837
+- Unique questions: 14,590
+- Sources: {'kaggle_v1': 9909, 'negative': 2500, 'hard_negative': 2500}
+
+## Fields
+| Field | Description |
+|-------|-------------|
+| `image` | PIL-ready surveillance image |
+| `messages` | Qwen chat JSON (user question + assistant thinking/answer) |
+| `ground_truth` | Pixel xyxy boxes + class |
+| `ground_truth_norm` | 0-1000 normalized boxes |
+| `categories` | QA category labels |
+| `dup_weight` | Sampling weight for deduplicated training |
+| `source` | kaggle_v1 / negative / hard_negative |
 
 ## Usage
-
 ```python
 from datasets import load_dataset
+import json
 
-ds = load_dataset("yunus-emre/uav-drone-reasoning-vqa")
-
-# Access a sample
-sample = ds["train"][0]
-image = sample["image"]  # PIL Image
-qa_pairs = json.loads(sample["conversations"])
-gt_boxes = json.loads(sample["ground_truth"])
-
-for qa in qa_pairs:
-    print(f"Q: {qa['question']}")
-    print(f"T: {qa['thinking']}")
-    print(f"A: {qa['answer']}")
-    print()
+ds = load_dataset("yunus-emre/uav-drone-reasoning-vqa", "v2")
+row = ds["train"][0]
+messages = json.loads(row["messages"])
 ```
-
-## License
-Apache 2.0