Conversation
…adjust SVG dimensions for consistency
…y accent handling in layout engine
…lignment specifications
…multi-dot accents
…h kerning option in layout options
…ages, and example inputs for improved user experience
Deploying ratex with
|
| Latest commit: |
25f6750
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://811b8e08.ratex.pages.dev |
| Branch Preview URL: | https://improve.ratex.pages.dev |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 25f67506c9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| mask = get_ink_mask(img) | ||
| if not np.any(mask): | ||
| return img[:10, :10] # fallback: tiny empty | ||
| return img[:10, :10] |
There was a problem hiding this comment.
Reject one-sided blank renders before shape scoring
When RaTeX emits an all-white PNG for a newly added, roughly square formula, this substitutes a 10×10 white crop; the later metric can then award up to 0.4 from aspect and width similarity alone, exceeding the default 0.30 pass threshold despite rendering no ink. Because the PNG is considered scored, coverage checks also accept it, and a new formula has no prior per-case baseline to catch the failure. Detect whether exactly one image has no ink and assign it a failing score, while handling intentionally blank reference/output pairs separately.
Useful? React with 👍 / 👎.
No description provided.