README.md

---
dataset_info:
  config_name: default
  features:
  - name: image
    dtype:
      image:
        decode: false
  - name: text
    dtype: string
  - name: line_id
    dtype: string
  - name: line_reading_order
    dtype: int64
  - name: line_coords
    dtype:
      sequence:
        sequence: int64
  - name: line_baseline
    dtype:
      sequence:
        sequence: int64
  - name: line_augmentation
    dtype: string
  - name: region_id
    dtype: string
  - name: region_reading_order
    dtype: int64
  - name: region_type
    dtype: string
  - name: region_coords
    dtype:
      sequence:
        sequence: int64
  - name: filename
    dtype: string
  - name: project_name
    dtype: string
  - name: inference_20260531_140505_448797_model_microsoft_trocr-small-printed
    dtype: string
  splits:
  - name: train
    num_examples: 604
    num_bytes: 7770494
  download_size: 7770494
  dataset_size: 7770494
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train/**/*.parquet
tags:
- image-to-text
- htr
- trocr
- inference
- pagexml
license: mit
---

# Dataset Card for inference-test-dataset

This dataset is derived from [danameyer/printed-test-lines](https://huggingface.co/datasets/danameyer/printed-test-lines) and has been enriched with inference results.

## Dataset Summary

This dataset contains 604 samples across 1 split(s).

### Projects Included

printed_test

## Duplicate Line Information

Duplicate line statistics are calculated from the dataset key columns `filename`, `region_id`, and `line_id`, plus `project_name` when project metadata is available.

Only original rows are counted here.

- Duplicate rows: 0
- Duplicate groups: 0
- Duplicate excess rows: 0

### Duplicate Lines by Split

- **train**: 0 duplicate rows, 0 duplicate groups, 0 duplicate excess rows

## Dataset Structure

### Data Splits

- **train**: 604 samples

### Dataset Size

- Approximate total size: 7.41 MB
- Total samples: 604

### Features

- **image**: `Image(mode=None, decode=False)`
- **text**: `Value('string')`
- **line_id**: `Value('string')`
- **line_reading_order**: `Value('int64')`
- **line_coords**: `List(List(Value('int64')))`
- **line_baseline**: `List(List(Value('int64')))`
- **line_augmentation**: `Value('string')`
- **region_id**: `Value('string')`
- **region_reading_order**: `Value('int64')`
- **region_type**: `Value('string')`
- **region_coords**: `List(List(Value('int64')))`
- **filename**: `Value('string')`
- **project_name**: `Value('string')`
- **inference_20260531_140505_448797_model_microsoft_trocr-small-printed**: `Value('string')`

## Data Organization

Data is organized as parquet shards by split and project:

```
data/
├── <split>/
│   └── <project_name>/
│       └── <timestamp>-<shard>.parquet
```

The HuggingFace Hub automatically merges all parquet files when loading the dataset.

## Usage

```python
from datasets import load_dataset

# Load entire dataset
dataset = load_dataset("danameyer/inference-test-dataset")

# Load specific split
dataset_split = load_dataset("danameyer/inference-test-dataset", split="train")
```