hoonisone/AtomicTags

An Atomic Multi-Tag Taxonomy for Fine-Grained Diagnosis of Scene Text Recognition Models

Atomic multi-tag annotations for K-STR-Bench and Union14M. One text crop can carry several tags, so a scene text recognition model can be diagnosed by the type of failure. Images are stored once. Tags are columns in the CSV. image_path is relative to the folder that contains the CSV.

Myounghun Han, Jin-Hyuk Hong · WACV 2027

Accepted to WACV 2027. The proceedings link is not available yet. This release is hoonisone/AtomicTags. Code and documentation are on GitHub.

Data Sources

K-STR-Bench. A challenge-aware benchmark for Korean scene text recognition. GitHub | Hugging Face. The split here is K_STR_Bench, from GIST_Signboard and KAIST_Corrected. AIHub_Standardized is part of K-STR-Bench and is not included. The K-STR-Bench release includes every crop and the columns is_only_korean and is_illegible for filtering. This release is already filtered: text.v1.0.csv and the image folders contain only crops where is_only_korean is true and is_illegible is false.

Union14M. A scene text recognition dataset collected from 14 public datasets (Jiang et al., ICCV 2023). GitHub. Union14M-L and Union14M-Benchmark are released from that repository. The splits here are Union14M_Benchmark and Union14M_General. Each CSV and its image folder already contain only the crops in this release. Tag columns in these tables are separate from the original Union14M annotations.

K_STR_Bench/
├── text.v1.0.csv
├── GIST_Signboard/text_images/
└── KAIST_Corrected/text_images/
Union14M_Benchmark/
├── text.v1.0.csv
└── text_images/
Union14M_General/
├── text_v.1.0.csv
└── text_images/

License

Source code is under the MIT License. That license does not cover the images or the annotations.

These files do not share one license. Union14M-General and Union14M-Benchmark are not separate licenses. An image keeps the license of the Union14M dataset it came from. Names and terms below follow the Union14M source table. Unknown means that table does not state a license.

Source	Subset	License / terms
K-STR-Bench	GIST_Signboard	CC BY 4.0
K-STR-Bench	KAIST_Corrected	CC BY-SA 3.0
K-STR-Bench	AIHub_Standardized	Not included. Not redistributed; AIHub Terms of Use
Union14M	KAIST	CC BY-SA 3.0
Union14M	NEOCR	CC BY-NC-SA 4.0
Union14M	Uber-Text	Unknown
Union14M	RCTW	Unknown
Union14M	IIIT-ILST	CC BY 4.0
Union14M	MTWI	CC BY-NC 4.0
Union14M	COCOTextV2	CC BY 4.0
Union14M	LSVT	Unknown
Union14M	MLT19	CC BY-NC 4.0
Union14M	ReCTS	Unknown
Union14M	ArT	Unknown
Union14M	IntelOCR	Apache License 2.0
Union14M	TextOCR	CC BY 4.0
Union14M	HierText	CC BY-SA 4.0

Statistics

Inclusive counts crops where that tag is true, including crops with other true tags. Exclusive counts crops whose only true tag is that tag. K-STR-Bench has no rotated tag.

Tag	K-STR-Bench		Union14M-General		Union14M-Benchmark
	Inclusive	Exclusive	Inclusive	Exclusive	Inclusive	Exclusive
Crops	20,086		9,994		7,888
Stylized	2,273	925	791	171	1,727	311
Inconsistent size	602	94	253	25	599	19
Incomplete	767	258	844	133	759	82
Low visibility	1,352	598	6,984	4,743	307	25
Noisy	2,844	1,541	696	109	1,540	143
Distracted	3,019	644	880	31	3,733	127
Perspective distortion	3,268	945	1,231	33	676	4
Non-single-line	118	34	0	0	0	0
Non-straight	479	79	155	6	2,731	203
Rotated	—		2,127	214	4,928	536

True tags on a crop	K-STR-Bench	Union14M-General	Union14M-Benchmark
0	10,941	1,389	745
1	5,118	5,465	1,450
2	2,814	1,685	2,671
3	942	852	2,100
4	216	467	738
5	46	115	156
6	7	20	21
7	2	1	6
8	0	0	1

Citation

If you use these annotations, please cite:

@inproceedings{atomicmultitag2027,
    title     = {An Atomic Multi-Tag Taxonomy for Fine-Grained Diagnosis of Scene Text Recognition Models},
    author    = {Han, Myounghun and Hong, Jin-Hyuk},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    year      = {2027}
}