README

README.md

@@ -1,3 +1,226 @@
----
-license: cc-by-nc-sa-4.0
----
+---
+license: cc-by-nc-sa-4.0
+tags:
+- math
+- synthetic
+- handwritten
+- formula-recognition
+task_categories:
+- image-to-text
+---
+
+# AmbHMEG: Ambiguous Handwritten Mathematical Expression Dataset
+
+This repository provides a dataset of ambiguous handwritten mathematical expressions generated by AmbHMEG, a model for rendering LaTeX expressions as visually ambiguous math images.
+It also includes the pretrained AmbHMEG model used for dataset generation.
+
+- GitHub: https://github.com/pfnet-research/AmbHMEG
+
+## Contents
+
+The resources are organized in the order of the data generation pipeline.
+
+### Training Dataset
+
+- `mathwriting_with_graph.tar.zst`
+
+A preprocessed version of the MathWriting dataset with graph annotations used for training the AmbHMEG model.
+
+### Pretrained Model
+
+- `gryphgen_graph_sd_mathwriting_epoch_50.pth`
+
+A model trained on the MathWriting dataset to generate handwritten mathematical expressions, including ambiguous variations.
+
+### Generated Dataset
+
+- `mathwriting_amb.tar.zst`
+
+A dataset of ambiguously generated handwritten mathematical expressions produced by AmbHMEG.
+
+#### Structure
+
+```text
+mathwriting_amb/
+  basic/
+    test/
+    valid/
+  layout/
+    test/
+    valid/
+  symbol/
+    test/
+    valid/
+````
+
+#### Subsets
+
+##### basic
+
+Standard single-conditioned generated dataset.
+
+##### layout
+
+Dual-conditioned dataset:
+
+* original expression
+* expression with superscripts and subscripts removed
+
+##### symbol
+
+Dual-conditioned dataset:
+
+* original expression
+* expression where symbols are replaced with visually similar alternatives
+
+###### Replacement Rules Used in Symbol Editing
+
+```python
+pairs = [
+    ("0", "6"),
+    ("0", "8"),
+    ("1", "7"),
+    ("1", "9"),
+    ("2", "7"),
+    ("3", "8"),
+    ("5", "6"),
+    ("5", "8"),
+    ("0", "O"),
+    ("0", "o"),
+    ("1", "l"),
+    ("1", "I"),
+    ("2", "z"),
+    ("2", "Z"),
+    ("5", "S"),
+    ("6", "b"),
+    ("8", "B"),
+    ("9", "g"),
+    ("t", "+"),
+    ("c", "e"),
+    ("u", "v"),
+    ("v", "r"),
+    ("m", "n"),
+    ("a", "d"),
+    ("i", "j"),
+    ("x", r"\times"),
+    ("O", "Q"),
+    ("I", "J"),
+    ("v", r"\nu"),
+    ("o", r"\omicron"),
+    ("p", r"\rho"),
+    ("n", r"\eta"),
+    ("u", r"\mu"),
+    ("a", r"\alpha"),
+    ("y", r"\gamma"),
+    ("2", r"\zeta"),
+    ("2", r"\gamma"),
+    ("x", r"\chi"),
+    ("w", r"\omega"),
+    ("e", r"\epsilon"),
+    (r"\epsilon", r"\in"),
+    (r"\phi", r"\varphi"),
+    (r"\theta", r"\vartheta"),
+    (r"\pi", r"\varpi"),
+    ("k", r"\kappa"),
+    (r"\sigma", "o"),
+    ("r", r"\gamma"),
+    ("z", r"\xi"),
+    (r"\beta", "B"),
+    ("-", r"\sim"),
+    (r"\cdot", "."),
+    ("1", "|"),
+    ("|", "l"),
+    (r"\sum", "E"),
+    (r"\prod", r"\Pi"),
+    (r"\partial", "d"),
+    (r"\infty", "8"),
+    (r"\subset", "c"),
+    (r"\in", r"\epsilon"),
+    (r"\subseteq", r"\subset"),
+    (r"\approx", r"\sim"),
+    (r"\equiv", "="),
+    ("0", r"\emptyset"),
+    (r"\emptyset", r"\phi"),
+    (r"\ell", "1"),
+    ("C", "c"),
+    ("I", "l"),
+    ("I", "i"),
+    ("K", "k"),
+    ("M", "m"),
+    ("N", "n"),
+    ("O", "o"),
+    ("P", "p"),
+    ("S", "s"),
+    ("U", "u"),
+    ("V", "v"),
+    ("W", "w"),
+    ("X", "x"),
+    ("Y", "y"),
+    ("Z", "z"),
+    ("I", "j"),
+    ("L", "l"),
+    ("Q", "q"),
+    ("R", "r"),
+    (r"\Gamma", r"\gamma"),
+    (r"\Delta", r"\delta"),
+    (r"\Theta", r"\theta"),
+    (r"\Lambda", r"\lambda"),
+    (r"\Xi", r"\xi"),
+    (r"\Pi", r"\pi"),
+    (r"\Sigma", r"\sigma"),
+    (r"\Phi", r"\phi"),
+    (r"\Psi", r"\psi"),
+    (r"\Omega", r"\omega"),
+    (r"\Sigma", "E"),
+    (r"\Pi", r"\prod"),
+]
+```
+
+#### Notes
+
+This dataset contains data derived from and includes original MathWriting samples. It is subject to the terms of the MathWriting dataset license (CC BY-NC-SA 4.0).
+
+## Data Format
+
+Each sample consists of an image and a corresponding label file.
+
+* Image: `<id>.png`
+* Label: `<id>.pkl` (JSON-serializable structure)
+
+## Usage
+
+### Download and Extract
+
+```bash
+tar --zstd -xvf *.tar.zst
+```
+
+### Training and Generation
+
+Refer to the official repository:
+
+[https://github.com/pfnet-research/AmbHMEG](https://github.com/pfnet-research/AmbHMEG)
+
+## License
+
+### Code and Model
+
+* Licensed under the MIT License
+
+The model is trained on the MathWriting dataset, and its use and outputs may be subject to the dataset license.
+
+### Dataset
+
+* Licensed under CC BY-NC-SA 4.0
+* MathWriting Dataset
+  [https://github.com/google-research/google-research/tree/master/mathwriting](https://github.com/google-research/google-research/tree/master/mathwriting)
+
+## Usage Restrictions
+
+* Commercial use is not allowed
+* Derivative works must be shared under CC BY-NC-SA 4.0
+* Attribution to the original MathWriting authors is required
+
+## Acknowledgements
+
+We thank the authors of the MathWriting dataset for providing this resource.