Model card
HTR-ConvText.
DAIR-Groupunknown65.9M paramsCNN+Transformer hybrid (ConvText block)
Handwritten Text Recognition model combining convolution and textual information. Uses a convolutional feature extractor with cross-attention over character embeddings to improve recognition of historical handwriting. 65.9M parameters.
§ 01 · Benchmarks
Every benchmark HTR-ConvText has a recorded score for.
| # | Benchmark | Area · Task | Metric | Value | Rank | Date | Source |
|---|---|---|---|---|---|---|---|
| 01 | read2016(line-level) | Computer Vision · Optical Character Recognition | test-wer | 15.7% | #5 | 2024-12-06 | source ↗ |
| 02 | read2016(line-level) | Computer Vision · Optical Character Recognition | test-cer | 3.6% | #6 | 2024-12-06 | source ↗ |
| 03 | IAM | Computer Vision · Handwriting Recognition | wer | 12.9% | #6 | — | source ↗ |
| 04 | lam(line-level) | Computer Vision · Optical Character Recognition | test-cer | 2.7% | #7 | 2024-12-06 | source ↗ |
| 05 | lam(line-level) | Computer Vision · Optical Character Recognition | test-wer | 7.0% | #7 | 2024-12-06 | source ↗ |
| 06 | IAM | Computer Vision · Handwriting Recognition | cer | 4.0% | #13 | — | source ↗ |
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 02 · Strengths by area
Where HTR-ConvText actually performs.
§ 03 · Papers
1 paper with results for HTR-ConvText.
§ 05 · Sources & freshness
Where these numbers come from.
arxiv
6
results
6 of 6 rows marked verified.