Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- table-question-answering
|
| 5 |
+
- image-to-text
|
| 6 |
+
tags:
|
| 7 |
+
- table-extraction
|
| 8 |
+
- benchmark
|
| 9 |
+
- fintabnet
|
| 10 |
+
- document-ai
|
| 11 |
+
- docld
|
| 12 |
+
pretty_name: DocLD FinTabNet Benchmark
|
| 13 |
+
size_categories:
|
| 14 |
+
- n<1K
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# DocLD FinTabNet Benchmark Results
|
| 18 |
+
|
| 19 |
+
Benchmark results for [DocLD](https://docld.com) table extraction on the
|
| 20 |
+
[FinTabNet](https://paperswithcode.com/dataset/fintabnet) dataset.
|
| 21 |
+
|
| 22 |
+
## Results Summary
|
| 23 |
+
|
| 24 |
+
| Metric | Value |
|
| 25 |
+
|--------|-------|
|
| 26 |
+
| **Mean Accuracy** | 80.6% |
|
| 27 |
+
| **Median** | 81.2% |
|
| 28 |
+
| **P25 / P75** | 75.2% / 85.8% |
|
| 29 |
+
| **Min / Max** | 49.2% / 100.0% |
|
| 30 |
+
| **Scored Samples** | 68 |
|
| 31 |
+
| **Total Samples** | 500 |
|
| 32 |
+
|
| 33 |
+
## Methodology
|
| 34 |
+
|
| 35 |
+
- **Dataset**: [FinTabNet_OTSL](https://huggingface.co/datasets/docling-project/FinTabNet_OTSL) — 500 samples from the test split
|
| 36 |
+
- **Extraction**: DocLD agentic table extraction (VLM-based, gpt-5-mini)
|
| 37 |
+
- **Scoring**: Needleman-Wunsch hierarchical alignment (same as [RD-TableBench](https://github.com/reductoai/rd-tablebench))
|
| 38 |
+
- **Output**: HTML tables with rowspan/colspan for merged cells
|
| 39 |
+
|
| 40 |
+
## Comparison
|
| 41 |
+
|
| 42 |
+
| Provider | FinTabNet Accuracy |
|
| 43 |
+
|----------|-------------------|
|
| 44 |
+
| **DocLD** | **80.6%** |
|
| 45 |
+
| GTE (IBM) | ~78% |
|
| 46 |
+
| TATR (Microsoft) | ~65% |
|
| 47 |
+
|
| 48 |
+
## Files
|
| 49 |
+
|
| 50 |
+
- `results.json` — Full benchmark results with per-sample scores
|
| 51 |
+
- `predictions/` — HTML predictions for each sample
|
| 52 |
+
- `charts/` — Visualization PNGs
|
| 53 |
+
|
| 54 |
+
## Links
|
| 55 |
+
|
| 56 |
+
- [DocLD](https://docld.com)
|
| 57 |
+
- [Blog Post](https://docld.com/blog/docld-fintabnet)
|
| 58 |
+
- [Benchmark Code](https://github.com/Doc-LD/fintabnet-bench)
|
| 59 |
+
- [RD-TableBench Results](https://docld.com/blog/docld-tablebench)
|