surenreddy/uvh-coco

---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*.parquet
- split: val
path: data/val-*.parquet
dataset_info:
features:
- name: image
dtype: image
- name: id
dtype: int32
- name: width
dtype: int32
- name: height
dtype: int32
- name: annotations
dtype: string
license: cc-by-4.0
task_categories:
- object-detection
tags:
- coco
- coco-80
- vehicle-detection
- indian-traffic
- auto-labeled
- uvh-26
size_categories:
- 10K<n<100K
---

# UVH-COCO Dataset

COCO-80 annotated version of the
[UVH-26](https://huggingface.co/datasets/iisc-aim/UVH-26) Indian traffic
surveillance dataset.

## Overview

| Property | Value |
|----------|-------|
| Source | surenreddy/uvh-26 (14 vehicle classes, ~26 K images) |
| Classes | 80 (standard COCO) |
| Vehicle annotations | Mapped from UVH-26 human labels (majority-voting) |
| Non-vehicle annotations | Auto-labeled: Grounding DINO + YOLOv8 fallback |
| Banner filter | Street banner/poster depictions suppressed or flagged |

## Annotation Fields

| Field | Type | Description |
|-------|------|-------------|
| category_id | int | COCO-80 class index (0-indexed) |
| bbox | [x,y,w,h] | Bounding box (COCO format, absolute px) |
| source | str | `"uvh"`, `"gdino"`, or `"yolo"` |
| uvh_class_id | int/null | Original UVH class ID (1-14) |
| uvh_class_name | str/null | Original UVH class name |
| confidence | float/null | Model confidence (null for UVH) |
| banner_status | str | `"keep"`, `"discard"`, or `"review"` |

## UVH → COCO Class Mapping

| UVH Classes | COCO Class |
|-------------|------------|
| Hatchback, Sedan, SUV, MUV, Three-wheeler | car |
| Bus, Mini-bus, Tempo-traveller | bus |
| Truck, LCV, Van, Other | truck |
| Two-wheeler | motorcycle |
| Bicycle | bicycle |

## Usage

```python
from datasets import load_dataset
ds = load_dataset("surenreddy/uvh-coco", split="train")
```

## Attribution

Based on UVH-26 by AIM @ IISc (CC BY 4.0).
Auto-labels generated with Grounding DINO and YOLOv8.