---
title: CalcTrainer Annotator
emoji: 📝
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.34.2
app_file: app.py
pinned: false
license: mit
short_description: Annotation tool for CalcTrainer handwriting dataset
dataset_info:
  features:
  - name: question_id
    dtype: string
  - name: handwriting_image
    dtype: image
  - name: math_problem
    dtype: string
  - name: correct_answer
    dtype: int64
  - name: ocr_prediction
    dtype: string
  - name: ocr_parsed_number
    dtype: int64
  - name: is_ocr_likely_correct
    dtype: bool
  - name: human_label
    dtype: string
  - name: was_corrected
    dtype: bool
  - name: annotation_status
    dtype: string
  splits:
  - name: train
    num_bytes: 4183724.125
    num_examples: 1447
  download_size: 4269963
  dataset_size: 4183724.125
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# CalcTrainer Annotator 📝

**Annotation tool for validating OCR predictions on handwritten math answers from [CalcTrainer](https://huggingface.co/spaces/hoololi/CalcTrainer)**

## How to Use

1. **Review** the math problem and handwritten answer
2. **Validate or correct** the OCR suggestion:
   - **✅ Correct** → Validates and moves to next
   - **❌ Incorrect** → Type correct number and validate
3. **Export** annotations anytime to save progress

## Features

- **Smart filtering**: Only shows unannotated entries
- **OCR suggestions**: Pre-filled predictions with error warnings
- **Incremental export**: Safely add annotations without data loss
- **Progress tracking**: Real-time statistics

## Data Flow

```
CalcTrainer → Raw Dataset → Manual Annotation → Validated Dataset
```

## Output

Creates `hoololi/CalcTrainer_dataset_annotated` with human-validated labels for OCR model fine-tuning.

**Private Space** - Controlled access for annotation quality.