Datasets:

thirdeyelabs
/

indian-road-dataset

@@ -3,12 +3,8 @@ license: cc-by-4.0
 task_categories:
 - object-detection
 - image-segmentation
-- other
 task_ids:
 - vehicle-detection
-language:
-- en
 tags:
 - autonomous-driving
 - indian-roads
@@ -16,9 +12,6 @@ tags:
 - bdd100k
 - computer-vision
 - detection
-- segmentation
-- tracking
-- gps
 pretty_name: Indian Road Driving Dataset
 size_categories:
 - 100K<n<1M
@@ -26,85 +19,71 @@ size_categories:
 # 🚗 Indian Road Driving Dataset
-<div align="center">
-**by [ThirdEye Labs](https://thirdeyelabs.ai)**
-[![Website](https://img.shields.io/badge/Website-thirdeyelabs.ai-blue?style=for-the-badge)](https://thirdeyelabs.ai)
-[![Demo](https://img.shields.io/badge/Live_Demo-View_Now-orange?style=for-the-badge)](https://thirdeyelabs.ai/demo)
-[![License](https://img.shields.io/badge/License-CC_BY_4.0-green?style=for-the-badge)](https://creativecommons.org/licenses/by/4.0/)
-*The largest open dataset of annotated Indian road footage — captured, processed, and released by ThirdEye Labs.*
-</div>
 ---
 ## 🌍 Why Indian Roads?
-Indian roads are among the most complex driving environments in the world — dense mixed traffic, unpredictable pedestrian behaviour, auto-rickshaws, cattle, informal lane usage, and extreme lighting conditions. Yet virtually no large-scale annotated dataset exists for this domain.
-Existing datasets (BDD100K, nuScenes, Waymo) are dominated by Western and East Asian road conditions. Models trained on them fail to generalise to India's 63 million vehicles and 1.4 billion people.
-**We built this dataset to change that.**
 ---
-## 📊 Dataset at a Glance
 | Metric | Value |
 |--------|-------|
-| Total clips | 8,441 |
-| Annotated frames | 646,014 |
-| Object detections | 6,896,202 |
-| Segmentation masks | 1,290,463 |
-| GPS-tagged frames | ✅ |
-| Annotation format | BDD100K |
-| Capture device | CP Plus dashcam |
-| Location | Delhi NCR, India |
-| Conditions | Day · Night · Dusk · Rain |
 ---
-## 🏷️ Detection Classes
-12 classes purpose-built for Indian roads:
-| Class | Description |
-|---|---|
-| `person` | Pedestrians |
-| `rider` | Motorcyclists / cyclists with rider |
-| `car` | Passenger cars |
-| `truck` | Trucks and tempos |
-| `bus` | Buses |
-| `motorcycle` | Motorcycles (unridden) |
-| `bicycle` | Bicycles |
-| `autorickshaw` | Auto-rickshaws (tuk-tuks) |
-| `animal` | Cattle, dogs, and other animals on road |
-| `vehicle fallback` | Unclassified vehicles |
-| `traffic light` | Traffic signals |
-| `traffic sign` | Road signs and boards |
 ---
 ## 📁 Dataset Structure
 ```
-indian-road-dataset/
-├── images/
-│   └── {clip_id}/
-│       ├── 0000.jpg
-│       ├── 0001.jpg
-│       └── ...
-├── annotations/
-│   ├── detection.json          # BDD100K format
-│   ├── tracking.json           # Multi-object tracking
-│   └── scene_attributes.json   # Weather, time of day, scene type
-├── masks/
-│   └── {clip_id}/
-│       └── {frame}.png         # Semantic segmentation masks
-└── gps/
-    └── gps_tracks.json         # GPS coordinates per clip
 ```
 ---
@@ -117,6 +96,8 @@ indian-road-dataset/
 from datasets import load_dataset
 ds = load_dataset("thirdeyelabs/indian-road-dataset")
 ```
 ### Load annotations directly
@@ -127,7 +108,7 @@ import json
 with open("annotations/detection.json") as f:
     annotations = json.load(f)
-# BDD100K format — each entry has:
 # { "name": "clip_id/frame", "labels": [{ "category": "car", "box2d": {...} }] }
 ```
@@ -139,9 +120,7 @@ huggingface-cli download thirdeyelabs/indian-road-dataset --repo-type dataset
 ---
-## 📐 Annotation Format
-Annotations follow the [BDD100K](https://doc.bdd100k.com/format.html) schema:
 ```json
 {
@@ -168,50 +147,32 @@ Annotations follow the [BDD100K](https://doc.bdd100k.com/format.html) schema:
 ## 🗺️ GPS Coverage
-Every clip is linked to GPS coordinates where available, enabling:
-- Geographic filtering by route / area
 - Speed and trajectory analysis
 - Map-based dataset exploration
 ---
-## 🏗️ How It Was Built
-This dataset was produced by the **ThirdEye Labs data pipeline** — an end-to-end ML annotation system built for scale:
-1. **Ingest** — raw MP4s from CP Plus dashcams uploaded to S3
 2. **Keyframe extraction** — 1 frame/second via FFmpeg
-3. **GPS parsing** — matched from companion `.srt` files
-4. **Object detection** — custom YOLO model fine-tuned on Indian road objects
-5. **Semantic segmentation** — SegFormer for drivable area and scene parsing
 6. **Multi-object tracking** — ByteTrack across frames
-7. **Scene classification** — weather, lighting, scene type per clip
-> Want to run your own data through this pipeline? [Contact us →](https://thirdeyelabs.ai/contact)
----
-## 📈 Benchmark Results
-Models fine-tuned on this dataset show significant improvements on Indian road scenarios. See our full benchmark on [thirdeyelabs.ai](https://thirdeyelabs.ai).
----
-## 🔗 Links
-| | |
-|---|---|
-| 🌐 **Website** | [thirdeyelabs.ai](https://thirdeyelabs.ai) |
-| 🎬 **Live Demo** | [thirdeyelabs.ai/demo](https://thirdeyelabs.ai/demo) |
-| 📧 **Contact** | [thirdeyelabs.ai/contact](https://thirdeyelabs.ai/contact) |
 ---
 ## 📜 License
-This dataset is released under the [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/) license.
-You are free to use, share, and adapt this data for any purpose — including commercial use — with attribution to **ThirdEye Labs (thirdeyelabs.ai)**.
 ---
@@ -229,10 +190,12 @@ You are free to use, share, and adapt this data for any purpose — including co
 ---
-<div align="center">
-  <a href="https://thirdeyelabs.ai">
-    <img src="https://thirdeyelabs.ai/og-image.png" alt="ThirdEye Labs" width="600"/>
-  </a>
-  *Built with ❤️ in India*
-</div>

 task_categories:
 - object-detection
 - image-segmentation
 task_ids:
 - vehicle-detection
 tags:
 - autonomous-driving
 - indian-roads
 - bdd100k
 - computer-vision
 - detection
 pretty_name: Indian Road Driving Dataset
 size_categories:
 - 100K<n<1M
 # 🚗 Indian Road Driving Dataset
+The **Indian Road Driving Dataset** is the largest open dataset of annotated Indian road footage, created by ThirdEye Labs. It addresses the critical gap in autonomous driving datasets for Indian road conditions.
 ---
 ## 🌍 Why Indian Roads?
+Indian roads present unique challenges absent from existing datasets (BDD100K, nuScenes, Waymo):
+- Dense mixed traffic with unpredictable behavior
+- Auto-rickshaws, cattle, and informal lane usage
+- Extreme lighting conditions
+- **63 million vehicles and 1.4 billion people** — yet no large-scale annotated dataset existed
 ---
+## 📊 Dataset Statistics
 | Metric | Value |
 |--------|-------|
+| **Total clips** | 8,441 |
+| **Annotated frames** | 646,014 |
+| **Object detections** | 6,896,202 |
+| **Segmentation masks** | 1,290,463 |
+| **GPS-tagged frames** | ✅ |
+| **Annotation format** | BDD100K |
+| **Capture device** | CP Plus dashcam |
+| **Location** | Delhi NCR, India |
+| **Conditions** | Day · Night · Dusk · Rain |
 ---
+## 🏷️ Detection Classes (12 classes)
+- **person** — Pedestrians
+- **rider** — Motorcyclists/cyclists with rider
+- **car** — Passenger cars
+- **truck** — Trucks and tempos
+- **bus** — Buses
+- **motorcycle** — Motorcycles (unridden)
+- **bicycle** — Bicycles
+- **autorickshaw** — Auto-rickshaws (tuk-tuks)
+- **animal** — Cattle, dogs, animals on road
+- **vehicle fallback** — Unclassified vehicles
+- **traffic light** — Traffic signals
+- **traffic sign** — Road signs and boards
 ---
 ## 📁 Dataset Structure
+Data is stored as **646 WebDataset tar shards** (`data/train-00000-of-00646.tar` … `data/train-00645-of-00646.tar`), each containing ~1,000 frames. Each frame has 3 files inside the shard:
 ```
+{clip_id}_{frame:04d}.jpg   # keyframe image
+{clip_id}_{frame:04d}.png   # segmentation mask
+{clip_id}_{frame:04d}.json  # BDD100K annotations (detections + scene attributes)
+```
+Standalone annotation files are also provided for convenient bulk access:
+```
+annotations/
+├── detection.json          # BDD100K format — all 646,014 frames (1.3 GB)
+└── scene_attributes.json   # per-clip weather, time of day, scene type
+gps/
+└── gps_tracks.json         # GPS coordinates per clip
 ```
 ---
 from datasets import load_dataset
 ds = load_dataset("thirdeyelabs/indian-road-dataset")
+sample = ds["train"][0]
+# sample keys: jpg, png, json
 ```
 ### Load annotations directly
 with open("annotations/detection.json") as f:
     annotations = json.load(f)
+# BDD100K format — each entry:
 # { "name": "clip_id/frame", "labels": [{ "category": "car", "box2d": {...} }] }
 ```
 ---
+## 📐 Annotation Format (BDD100K Schema)
 ```json
 {
 ## 🗺️ GPS Coverage
+Every clip includes GPS coordinates, enabling:
+- Geographic filtering by route/area
 - Speed and trajectory analysis
 - Map-based dataset exploration
 ---
+## 🏗️ Production Pipeline
+ThirdEye Labs end-to-end ML annotation system:
+1. **Ingest** — raw MP4s from CP Plus dashcams to S3
 2. **Keyframe extraction** — 1 frame/second via FFmpeg
+3. **GPS parsing** — matched from `.srt` files
+4. **Object detection** — custom YOLO fine-tuned for Indian roads
+5. **Semantic segmentation** — SegFormer for drivable areas
 6. **Multi-object tracking** — ByteTrack across frames
+7. **Scene classification** — weather, lighting, scene type
 ---
 ## 📜 License
+**Creative Commons Attribution 4.0 International (CC BY 4.0)**
+Free to use, share, and adapt for any purpose (including commercial) with attribution to **ThirdEye Labs**.
 ---
 ---
+## 🔗 Links
+- 🌐 **Website**: [thirdeyelabs.ai](https://thirdeyelabs.ai)
+- 🎬 **Demo**: [thirdeyelabs.ai/demo](https://thirdeyelabs.ai/demo)
+- 📧 **Contact**: [thirdeyelabs.ai/contact](https://thirdeyelabs.ai/contact)
+---
+*Built with ❤️ in India*