---
license: cc-by-4.0
task_categories:
  - object-detection
tags:
  - stamp-detection
  - document-analysis
  - object-detection
  - yolo
  - computer-vision
pretty_name: Clean Core Stamp Detection Dataset
size_categories:
  - 1K<n<10K
dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype: string
  splits:
    - name: train
      num_examples: 6283
    - name: validation
      num_examples: 785
    - name: test
      num_examples: 786
---

# Clean Core Stamp Detection Dataset

A cleaned document-stamp detection dataset built from curated raw sources only.
This version replaces the previous `mapo80/stamps` release and removes noisy or
off-target sources such as postage stamps, shape/identity datasets, corrupted
multi-class remaps, and duplicated negatives.

## Overview

| Parameter | Value |
|-----------|-------|
| Task | Object detection |
| Classes | 1 (`stamp`, class id 0) |
| Total images | 7854 |
| Positive images | 6608 |
| Negative images | 1246 |
| Total bounding boxes | 12659 |
| Annotation format | YOLO txt (`class x_center y_center width height`) |

## Splits

| Split | Total Images | Positive | Negative | Bounding Boxes |
|-------|-------------:|---------:|---------:|---------------:|
| Train | 6283 | 5287 | 996 | 10046 |
| Val | 785 | 660 | 125 | 1298 |
| Test | 786 | 661 | 125 | 1315 |

## What Changed

- Kept only in-scope document stamp sources
- Filtered `stamp_detection_stampa` to the raw `stamp` class only
- Removed postage-stamp, shape, identity, and corrupted mixed-class sources
- Rebuilt negatives from `RVL-CDIP` and `FUNSD`
- Dropped severe negative outliers (blank pages, almost-black pages, heavily degraded scans)
- Rebuilt train/val/test from scratch from `data/raw`

## Included Sources

### Positive sources

- `stamp_detection_jsam`: 4385 images
- `yolo_stamp_classify`: 1397 images
- `stamp_detectation_marcos`: 365 images
- `stamp_detection_stampa` (filtered): 461 images

### Negative sources

- `neg_rvl_cdip`: 1048 images
- `neg_funsd`: 198 images

## Excluded Source Families

- `detect_postage_stamp`
- `stamp_class`
- `stamp_shape`
- `stamp_individual`
- `stamp_recognition`
- `stamp_detection_shujing`
- `stamp_warisara`
- `staver_yolo`
- `neg_tobacco3482`

## Directory Structure

The repository stores the dataset as zip archives:

```text
train.zip
val.zip
test.zip
dataset.yaml
```

Each zip contains:

```text
images/<split>/
labels/<split>/
```

## `dataset.yaml`

```yaml
path: .
train: images/train
val: images/val
test: images/test

nc: 1
names:
  - stamp
```

## Notes

- Empty label files are valid negatives
- This benchmark is intentionally narrower and cleaner than the previous release
- It is optimized for document-level stamp detection, not postage stamps or stamp classification