feat: Add dataset configurations for eval.yaml
Browse files
eval.yaml
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
name: ArenaOCR
|
| 2 |
+
description: ArenaOCR is an extremely challenging document OCR benchmark featuring 5,000 unique, procedurally generated PDF pages and unit test cases. It covers six core categories of complex OCR challenges including academic arXiv mathematics, dense and multi-column tables, complex header/footer metadata, multi-column reading orders, degraded photocopy-style old scans, and dense microscopic text.
|
| 3 |
+
evaluation_framework: arenaocr
|
| 4 |
+
|
| 5 |
+
tasks:
|
| 6 |
+
- id: ArenaOCR_ArXivMath
|
| 7 |
+
config: default
|
| 8 |
+
split: arxiv_math
|
| 9 |
+
- id: ArenaOCR_HeadersFooters
|
| 10 |
+
config: default
|
| 11 |
+
split: headers_footers
|
| 12 |
+
- id: ArenaOCR_TableTests
|
| 13 |
+
config: default
|
| 14 |
+
split: table_tests
|
| 15 |
+
- id: ArenaOCR_MultiColumn
|
| 16 |
+
config: default
|
| 17 |
+
split: multi_column
|
| 18 |
+
- id: ArenaOCR_OldScans
|
| 19 |
+
config: default
|
| 20 |
+
split: old_scans
|
| 21 |
+
- id: ArenaOCR_LongTinyText
|
| 22 |
+
config: default
|
| 23 |
+
split: long_tiny_text
|