Datasets:
Add image-to-text task category and link paper
#2
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,10 +1,12 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
|
| 5 |
<h1 align="center">MORE: A <u><strong>M</strong></u>ultilingual D<u><strong>o</strong></u>cument Pa<u><strong>r</strong></u>sing B<u><strong>e</strong></u>nchmark and Evaluation</h1>
|
| 6 |
|
| 7 |
-
<p align="center"><nobr><a href="https://openreview.net/pdf?id=ov240fehF6"><img alt="Paper: ICML2026" src="https://img.shields.io/badge/Paper-ICML2026-b31b1b.svg"></a> <a href="https://github.com/zimoqingfeng/MORE"><img alt="GitHub: MORE" src="https://img.shields.io/badge/GitHub-MORE-181717.svg?logo=github"></a></nobr></p>
|
| 8 |
|
| 9 |
<p align="center">
|
| 10 |
<a href="#overview">π Overview</a> |
|
|
@@ -333,39 +335,39 @@ MORE follows the OmniDocBench v1.5 evaluation protocol and extends it to code an
|
|
| 333 |
|
| 334 |
Sequential tasks use normalized edit distance:
|
| 335 |
|
| 336 |
-
|
| 337 |
\begin{aligned}
|
| 338 |
\mathrm{NED}
|
| 339 |
&= 1 - \frac{1}{N} \sum_{i=1}^{N}
|
| 340 |
\frac{\mathrm{EditDist}(P_i, G_i)}
|
| 341 |
{\max(\lvert P_i\rvert, \lvert G_i\rvert)}
|
| 342 |
\end{aligned}
|
| 343 |
-
|
| 344 |
|
| 345 |
Formula recognition uses Character Detection Matching:
|
| 346 |
|
| 347 |
-
|
| 348 |
\begin{aligned}
|
| 349 |
\mathrm{CDM}
|
| 350 |
&= \frac{1}{N} \sum_{i=1}^{N}
|
| 351 |
\frac{2 \cdot TP_i}{2 \cdot TP_i + FP_i + FN_i}
|
| 352 |
\end{aligned}
|
| 353 |
-
|
| 354 |
|
| 355 |
Table recognition uses TEDS:
|
| 356 |
|
| 357 |
-
|
| 358 |
\begin{aligned}
|
| 359 |
\mathrm{TEDS}
|
| 360 |
&= 1 - \frac{1}{N} \sum_{i=1}^{N}
|
| 361 |
\frac{\mathrm{TreeEditDist}(T_{p,i}, T_{g,i})}
|
| 362 |
{\max(\lvert T_{p,i}\rvert, \lvert T_{g,i}\rvert)}
|
| 363 |
\end{aligned}
|
| 364 |
-
|
| 365 |
|
| 366 |
The overall score is the arithmetic mean of the six task scores:
|
| 367 |
|
| 368 |
-
|
| 369 |
\begin{aligned}
|
| 370 |
\mathrm{Overall}
|
| 371 |
&= \frac{1}{6}
|
|
@@ -375,7 +377,7 @@ $$
|
|
| 375 |
+ \sum_{k \in \mathcal{S}} \mathrm{NED}_k
|
| 376 |
\right)
|
| 377 |
\end{aligned}
|
| 378 |
-
|
| 379 |
|
| 380 |
<a id="leaderboard"></a>
|
| 381 |
|
|
@@ -449,4 +451,4 @@ We sincerely hope MORE is useful for your research. If it helps your work, we wo
|
|
| 449 |
booktitle = {International Conference on Machine Learning},
|
| 450 |
year = {2026}
|
| 451 |
}
|
| 452 |
-
```
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- image-to-text
|
| 5 |
---
|
| 6 |
|
| 7 |
<h1 align="center">MORE: A <u><strong>M</strong></u>ultilingual D<u><strong>o</strong></u>cument Pa<u><strong>r</strong></u>sing B<u><strong>e</strong></u>nchmark and Evaluation</h1>
|
| 8 |
|
| 9 |
+
<p align="center"><nobr><a href="https://huggingface.co/papers/2607.02956"><img alt="Paper" src="https://img.shields.io/badge/HF-Paper-blue.svg"></a> <a href="https://openreview.net/pdf?id=ov240fehF6"><img alt="Paper: ICML2026" src="https://img.shields.io/badge/Paper-ICML2026-b31b1b.svg"></a> <a href="https://github.com/zimoqingfeng/MORE"><img alt="GitHub: MORE" src="https://img.shields.io/badge/GitHub-MORE-181717.svg?logo=github"></a></nobr></p>
|
| 10 |
|
| 11 |
<p align="center">
|
| 12 |
<a href="#overview">π Overview</a> |
|
|
|
|
| 335 |
|
| 336 |
Sequential tasks use normalized edit distance:
|
| 337 |
|
| 338 |
+
```math
|
| 339 |
\begin{aligned}
|
| 340 |
\mathrm{NED}
|
| 341 |
&= 1 - \frac{1}{N} \sum_{i=1}^{N}
|
| 342 |
\frac{\mathrm{EditDist}(P_i, G_i)}
|
| 343 |
{\max(\lvert P_i\rvert, \lvert G_i\rvert)}
|
| 344 |
\end{aligned}
|
| 345 |
+
```
|
| 346 |
|
| 347 |
Formula recognition uses Character Detection Matching:
|
| 348 |
|
| 349 |
+
```math
|
| 350 |
\begin{aligned}
|
| 351 |
\mathrm{CDM}
|
| 352 |
&= \frac{1}{N} \sum_{i=1}^{N}
|
| 353 |
\frac{2 \cdot TP_i}{2 \cdot TP_i + FP_i + FN_i}
|
| 354 |
\end{aligned}
|
| 355 |
+
```
|
| 356 |
|
| 357 |
Table recognition uses TEDS:
|
| 358 |
|
| 359 |
+
```math
|
| 360 |
\begin{aligned}
|
| 361 |
\mathrm{TEDS}
|
| 362 |
&= 1 - \frac{1}{N} \sum_{i=1}^{N}
|
| 363 |
\frac{\mathrm{TreeEditDist}(T_{p,i}, T_{g,i})}
|
| 364 |
{\max(\lvert T_{p,i}\rvert, \lvert T_{g,i}\rvert)}
|
| 365 |
\end{aligned}
|
| 366 |
+
```
|
| 367 |
|
| 368 |
The overall score is the arithmetic mean of the six task scores:
|
| 369 |
|
| 370 |
+
```math
|
| 371 |
\begin{aligned}
|
| 372 |
\mathrm{Overall}
|
| 373 |
&= \frac{1}{6}
|
|
|
|
| 377 |
+ \sum_{k \in \mathcal{S}} \mathrm{NED}_k
|
| 378 |
\right)
|
| 379 |
\end{aligned}
|
| 380 |
+
```
|
| 381 |
|
| 382 |
<a id="leaderboard"></a>
|
| 383 |
|
|
|
|
| 451 |
booktitle = {International Conference on Machine Learning},
|
| 452 |
year = {2026}
|
| 453 |
}
|
| 454 |
+
```
|