Skip to content

Automatically export evaluation results as markdown tables

Depends #236 (closed) #237 (closed)

We want to export the results so that these tables are automatically filled.

The idea is to compute a table with the following columns, for each split. This should be a new function in dan.ocr.utils, called right after evaluation. The evaluate function should return the metrics so you can use them in dan.ocr.evaluate.

  • "Split", one of train/val/test
  • "CER (HTR-NER)" = "cer"
  • "CER (HTR)" = "cer_no_tokens"
  • "WER (HTR-NER)" = "wer"
  • "WER (HTR)" = "wer_no_tokens"
  • "WER (HTR no punct)" = "wer_no_punct"
  • "NER" = "ner"

You need to create the table then add one row at a time.

Edited by Yoann Schneider