jpnadas/finrod
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
74b88775508983fa12387fd7321b6cc01704d3d1
finrod/CHANGELOG.md
João Pedro Battistella Nadas 74b8877550 feat(ocr): crop + rotate modal before recognition 
After picking an image, a modal opens with cropperjs's drag-handle
crop UI and Rotate left / Rotate right / Reset buttons. Cropping
to just the text region is the single biggest quality win for
OCR on phone photos — strips out logos, lighting transitions, and
background junk that bleed past the confidence filter.

- Lazy-loads cropperjs + its CSS from esm.sh on first use (no
  bundling; same pattern as Tesseract.js and CodeMirror).
- Honours EXIF orientation from phone shots (checkOrientation).
- Keyboard: Esc cancels, Enter accepts. Clicking the backdrop
  also cancels.
- Output goes through the existing preprocessForOCR pipeline
  (downscale to 1600px + Bradley adaptive threshold) before
  Tesseract sees it.
2026-06-09 19:02:51 +02:00

4.5 KiB
Raw Blame History

Changelog

All notable changes to finrod are documented here.

Format: Keep a Changelog. Versioning: Semantic Versioning.

[Unreleased]

Added

  • "Scan image" button next to the input pane: opens the phone's rear camera (or a file picker on desktop), runs OCR locally in the browser via Tesseract.js (loaded from esm.sh on first use), and replaces the editor contents with the recognised text. Language follows the "From" dropdown via an ISO 639-2 mapping; "Auto-detect" loads a common European set (eng+por+deu+nld). Progress status shows beneath the input ("Loading OCR engine…" → "OCR 40%" → "OCR done in 3.2s"). Zero image-size cost — same CDN pattern as CodeMirror.
  • Images are downscaled to 1600 px on the long edge before OCR. A 12 MP phone photo (~4000 px) drops to ~1.3 MP, making Tesseract 4-8× faster with no real accuracy loss for printed text. Smaller-than-1600 px inputs are passed through unchanged.
  • Adaptive thresholding (Bradley's method with integral images) binarises the downscaled image before OCR — drops hallucinated text from shadows, gradients, and surface texture on phone photos. Output is pure black-on-white, much closer to the scanned-page input Tesseract was trained on.
  • Per-word confidence filter (≥ 60) drops the low-confidence noise Tesseract still emits. Re-assembles lines from kept words; status line reports how many words were dropped.
  • Crop & rotate modal between image pick and OCR. After choosing an image, a modal pops up with cropperjs's drag-handle crop UI and Rotate left / Rotate right / Reset buttons. EXIF orientation is honoured automatically. Cancel / Esc / clicking the backdrop aborts; Enter accepts. Library lazy-loaded from esm.sh (no bundling).

Fixed

  • "Hot: …" status no longer claims "unloads in X" when the server's keep_alive is -1 (Ollama returns an absurd year-4001 expires_at). Threshold: any expiry >7 days drops the suffix and just shows the model name.

[0.2.0] - 2026-06-09

Added

  • Vim bindings in the input via CodeMirror 6 + @replit/codemirror-vim (loaded from esm.sh, no build step). Supports i/Esc, hjkl, word/line motions, dd/yy/p (vim's internal register), undo/redo, / search, : command line. Auto-disabled on touch devices (detected via (pointer: coarse)) where modal editing fights the virtual keyboard.
  • Shift-H / Shift-L cycle focus across the page controls (source → target → model → editor → button) like Shift-Tab / Tab. Also overridden in vim normal mode (replacing vim's viewport-top/bottom default) so the same shortcut works to leave the editor.
  • On page load, the model dropdown defaults to whichever model Ollama already has hot — skips the 5-15 s disk-reload pain on the first translation. After load it follows the user's manual selection.

Changed

  • Editor font-size pinned to 1rem to match the output pane (CodeMirror 6 ships smaller by default).

[0.1.0] - 2026-06-09

First release. Stateless translator UI that proxies streaming requests to Ollama on radagast.

Added

  • FastAPI backend with three endpoints:
    • GET /api/models — filtered (gemma,aya by default) list from Ollama's /api/tags
    • POST /api/translate — streams Ollama's NDJSON /api/generate response back to the browser
    • GET /api/loaded — proxies Ollama's /api/ps for the currently-loaded model
    • GET /healthz — for Kubernetes probes
  • Single-page UI:
    • Source / target / model dropdowns (30+ languages, "Auto-detect" source)
    • Token-by-token streaming into the output pane with a blinking caret
    • Live status line: spinner while waiting for first token, pulsing dot during streaming, elapsed time and token count
    • Glowing border on the output pane during streaming
    • Loaded-model indicator (● prefix in dropdown, "Hot: … — unloads in Xm" status line); refreshes on page load, on translation completion, and 500 ms after typing pauses
    • ⌘/Ctrl-Enter to translate from the input
    • Dark mode, inline SVG favicon (Star of Fëanor)
  • Dockerfile: two-stage uv build, slim Python 3.13 runtime
  • build.sh: multi-arch (amd64 + arm64) manual push to Scaleway, tags :latest, :<git-short-sha>, and :<pyproject-version>
  • Cluster deploy via the valinor repo at apps/finrod/ (bjw-s app-template, internal-only ingress at finrod.jpnadas.xyz)

Fixed

  • Lock against public PyPI explicitly — the workstation's global uv config had pointed uv.lock sources at the dexterenergy private artifact registry, which 401'd inside Docker