README.md

Process data from paperswithcode

See https://huggingface.co/datasets/pwc-archive/files/tree/main.

Download and unzip evaluation tables:

curl -L -O "https://huggingface.co/datasets/pwc-archive/files/resolve/main/jul-28-evaluation-tables.json.gz"
gunzip jul-28-evaluation-tables.json.gz

Install jq. See https://jqlang.org/. If on Debian/Ubuntu, install with sudo apt-get install jq.

Example jq to extract:

jq -r '
  def process(parent):
    .task as $current_task |
    (if parent then parent + " > " + $current_task else $current_task end) as $full_path |
    (.datasets[]? |
      .dataset as $dataset |
      .sota.rows[]? |
      {
        task_path: $full_path,
        dataset: $dataset,
        model_name: .model_name,
        paper_url: .paper_url,
        metrics: .metrics
      }
    ),
    (.subtasks[]? | process($full_path));
  
  ["task_path", "dataset", "model_name", "paper_url", "metric_name", "metric_value"],
  (
    [.[] | process(null)] |
    .[] |
    [.task_path, .dataset, .model_name, .paper_url] + 
    (.metrics | to_entries[] | [.key, .value]) |
    flatten
  ) |
  @csv
' jul-28-evaluation-tables.json > results.csv

Should get 326,393 rows in results.csv and looks like this:

~/paperswithcode-data> nu -c "open results.csv | length"
# 326393
~/paperswithcode-data> nu -c "open results.csv | skip 100 | take 10"
# ╭───┬────────────────────────────────────────────────────────────────────┬─────────────────┬───────────────┬────────────────────────────────────┬─────────────┬──────────────╮
# │ # │                             task_path                              │     dataset     │  model_name   │             paper_url              │ metric_name │ metric_value │
# ├───┼────────────────────────────────────────────────────────────────────┼─────────────────┼───────────────┼────────────────────────────────────┼─────────────┼──────────────┤
# │ 0 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ HTR-VT        │ https://arxiv.org/abs/2409.08573v1 │ Test CER    │         2.80 │
# │ 1 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ HTR-VT        │ https://arxiv.org/abs/2409.08573v1 │ Test WER    │         7.40 │
# │ 2 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-24 │ https://arxiv.org/abs/2006.07491v1 │ Test CER    │         3.00 │
# │ 3 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-24 │ https://arxiv.org/abs/2006.07491v1 │ Test WER    │        11.00 │
# │ 4 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-18 │ https://arxiv.org/abs/2006.07491v1 │ Test CER    │         3.10 │
# │ 5 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-18 │ https://arxiv.org/abs/2006.07491v1 │ Test WER    │        11.10 │
# │ 6 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-12 │ https://arxiv.org/abs/2006.07491v1 │ Test CER    │         3.10 │
# │ 7 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ OrigamiNet-12 │ https://arxiv.org/abs/2006.07491v1 │ Test WER    │        11.20 │
# │ 8 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ TrOCR         │ https://arxiv.org/abs/2109.10282v5 │ Test CER    │         3.60 │
# │ 9 │ Optical Character Recognition (OCR) > Handwritten Text Recognition │ LAM(line-level) │ TrOCR         │ https://arxiv.org/abs/2109.10282v5 │ Test WER    │        11.60 │
# ╰───┴────────────────────────────────────────────────────────────────────┴─────────────────┴───────────────┴────────────────────────────────────┴─────────────┴──────────────╯