---
license: mit
task_categories:
- object-detection
- instance-segmentation
language:
- en
tags:
- roboflow
- coco
- instance-segmentation
size_categories:
- 100K<n<1M
---

# Project Nova Merged — Version 3

Exported from [Roboflow](https://roboflow.com) workspace `project-nova-sp0ia`, project `project-nova-merged-d97ey`, **version 3**.

## Dataset Info

| Split | Images |
|-------|--------|
| Train | 126,988 |
| Valid | 7,906 |
| Test  | 7,829 |
| **Total** | **142,723** |

- **Task**: Instance Segmentation (COCO format)
- **Image size**: 640×640 (Fit with black edges)
- **Preprocessing**: Auto-orient, Adaptive Equalization contrast, filter null ≥100%
- **Augmentations**: Horizontal flip, ±20% brightness, 15° rotation, blur 1.5px, 2% noise, ±15% saturation, mosaic (×2 output versions)

## Files

| File | Description |
|------|-------------|
| `data.zip` | Full dataset (images + annotations, COCO JSON format) |
| `train/_annotations.coco.json` | COCO annotations for the training split |
| `valid/_annotations.coco.json` | COCO annotations for the validation split |
| `test/_annotations.coco.json` | COCO annotations for the test split |

## Usage

Download `data.zip` and unzip to access all images organized in `train/`, `valid/`, and `test/` folders, each containing a `_annotations.coco.json` file.