Sakana AIMetom
Metom
Metom is a Vision Transformer classifier for kuzushiji. It takes an image containing one character and returns a character label or ranked candidates.
Research release
The input is one character, not an entire page
The card describes separate training, validation and evaluation splits and supplies top-k label examples with probabilities. Page layout analysis and character segmentation are outside the stated classifier input.Metom official model card
Working with this release
Practical guidance. Crop a single character before classification. Preserve several candidates for damaged or ambiguous marks so a human can compare them with surrounding text.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/Metom |
| Training scope | 2,703 character classes, each occurring at least five times in the source dataset.Metom official model card |
| Training data | ROIS-DS CODH Japanese classical-book character dataset.Metom official model card |
| License | Apache 2.0Metom official model card |