Skip to content

Releases: naeyn/de-htr-web

DE·HTR v0.2.0 — de-htr-web-v2 model, config-driven multi-model support

Choose a tag to compare

@naeyn naeyn released this 01 Aug 14:56

Ships the de-htr-web-v2 model as the default recognizer, with v1 selectable in-app for comparison.

Highlights

  • v2 model default: 8.01% CER / 32.3% WER on the sealed writer-disjoint ScaDS test split (v1: 8.61% / 33.9%), greedy CTC at batch 1 — the deployment configuration.
  • Config-driven model contracts (format_version): v1 keeps its RGB + external ImageNet normalization path byte-identical; v2 uses 1-channel grayscale, [0,1] scaling, in-model normalization, raw-logits output, 112-symbol alphabet, max width 1024.
  • Model picker UI — switch v1/v2 live and re-run on the loaded image.
  • Fix: v2 emits batch-major [1, T, C] (v1 is time-major [T, 1, C]); decode verified bit-exact against the Python reference pipeline.

Artifacts

Scope

Modern German handwriting, one pre-cropped line; no page segmentation, no historical scripts (Kurrent/Sütterlin). Weights Apache-2.0; training-data attribution in the model repo's NOTICE-training-data.md.

DE·HTR v0.1.0 — line-level browser release

Choose a tag to compare

@naeyn naeyn released this 29 Jul 19:57

Modern German handwriting recognition for pre-cropped single lines.\n\n- Browser demo: https://de-htr-web.vercel.app\n- Complete model repository, PyTorch checkpoint, results, and training manifest: https://huggingface.co/naeyn/de-htr-web\n- int8 ONNX: 9,843,872 bytes, SHA-256 05cce04d3938a64edaee79caf2a74587eaa9f8c46c146d6e04f0512bfce51d82\n- Browser median inference: 259 ms on Apple M1 / Chrome 151\n- Scope: modern German handwriting, one pre-cropped line; no page segmentation\n\nLicensed under Apache-2.0. Source dataset and font attributions are included in the model repository.