Releases · lukas-blecher/LaTeX-OCR (original) (raw)

0.0.31

What's Changed

Big change (maybe breaking in your old environment) is moving away from PyQt5 to PyQt6. This will help with installation issues for M1 and M2 based systems.

• Update GUI to PyQt6 (Fix for M1 and M2 OS X) by @JoepdeJong in #236
• feat: add utf-8 support for reading README file by @llxlr in #201

New Contributors

• @llxlr made their first contribution in #201
• @JoepdeJong made their first contribution in #236

Full Changelog: 0.0.29...0.0.31

0.0.29

0.0.26

What's Changed

Mainly improved training

• add micro batches by @lukas-blecher in #148
• Data parallelism + pure ViT working by @TITC in #150

Full Changelog: 0.0.24...0.0.26

Breaking Changes

models.py moved into new module models.
Model class has forward moved to generate.

0.0.24

0.0.20

What's Changed

• Add Api by @lukas-blecher in #140
  The new functionality is built on new dependenciew. They can be installed by using the [api] or [all] tag when installing pix2tex[tag]
• LaTeX-OCR/dataset to LaTeX-OCR/pix2tex/dataset by @TITC in #123
• import problem by @TITC in #125
• preprocessing file missing by @TITC in #135

Breaking Changes

Methods in cli.py initialize and call_model have been moved into a the class LatexOCR.

Full Changelog: 0.0.4...0.0.20

0.0.4

What's Changed

• Add desktop entry script for the GUI by @JCGoran in #34
• 🔨 fix the image resize in pix2tex.call_model by @rainyl in #105
• alpha bug fixes by @TITC in #112
• [Feature] Add a cmdline option to use gnome-screenshot instead of pyqt to capture screenshots by @kxxt in #111
• img has been cut off abnormally by @TITC in #119
• Install via PyPI by @lukas-blecher in #121

New Contributors

• @JCGoran made their first contribution in #34
• @rainyl made their first contribution in #105
• @TITC made their first contribution in #112
• @kxxt made their first contribution in #111
• @frankier made their first contribution in #94

Full Changelog: v0.0.1...0.0.4

No new model checkpoints. See first release.

Weights release

Transitioning the weights from Google Drive to GitHub.

• weights contains the main network weights
• image_resizer contains the weights for the resizing network (optional)

LICENSE
Since the model was trained on arxiv data the weights are released under CC BY-NC-SA.