Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
44
84
GIST_Signboard/text_images/0/gist_3_1.jpg ์‹ ์˜์ฆ๊ถŒ
GIST_Signboard/text_images/0/gist_4_3.jpg ํ˜„๋Œ€๋ฆฌ๋ฐ”ํŠธ
GIST_Signboard/text_images/0/gist_4_4.jpg ๊ด‘์ฃผ์ „์‹œ์žฅ
GIST_Signboard/text_images/0/gist_7_1.jpg ์บกํ‹ด์•ˆ๋งˆ
GIST_Signboard/text_images/0/gist_7_2.jpg ๋ฒ„๋””
GIST_Signboard/text_images/0/gist_7_4.jpg ์ด๊ธ€
GIST_Signboard/text_images/0/gist_7_5.jpg ํ”ผํ”Œ
GIST_Signboard/text_images/0/gist_7_6.jpg ๋…ธ๋ž˜ํƒ€์šด
GIST_Signboard/text_images/0/gist_7_7.jpg ์ง€ํ•˜
GIST_Signboard/text_images/0/gist_8_2.jpg ํ”ผํ”Œ
GIST_Signboard/text_images/0/gist_8_3.jpg ๋…ธ๋ž˜ํƒ€์šด
GIST_Signboard/text_images/0/gist_8_4.jpg ์ง€ํ•˜
GIST_Signboard/text_images/0/gist_9_1.jpg ์บกํ‹ด์•ˆ๋งˆ
GIST_Signboard/text_images/0/gist_9_2.jpg ๋ฒ„๋””
GIST_Signboard/text_images/0/gist_9_4.jpg ์ด๊ธ€
GIST_Signboard/text_images/0/gist_10_2.jpg ์ค‘๋ฌธ์†ฅ๋šœ๊ป‘
GIST_Signboard/text_images/0/gist_10_3.jpg ์ œ์ฃผํ‘๋ผ์ง€์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_10_4.jpg ๊ด‘์ฃผ์ƒ๋ฌด์ 
GIST_Signboard/text_images/0/gist_11_1.jpg ์บกํ‹ด์•ˆ๋งˆ
GIST_Signboard/text_images/0/gist_12_1.jpg ์บกํ‹ด์•ˆ๋งˆ
GIST_Signboard/text_images/0/gist_12_5.jpg ์ค‘๋ฌธ์†ฅ๋šœ๊ป‘
GIST_Signboard/text_images/0/gist_12_6.jpg ์ œ์ฃผํ‘๋ผ์ง€์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_12_7.jpg ๊ด‘์ฃผ์ƒ๋ฌด์ 
GIST_Signboard/text_images/0/gist_12_9.jpg ํ”ผํ”Œ
GIST_Signboard/text_images/0/gist_12_10.jpg ๋…ธ๋ž˜ํƒ€์šด
GIST_Signboard/text_images/0/gist_12_12.jpg ์ง€ํ•˜
GIST_Signboard/text_images/0/gist_13_1.jpg ์ฒด๋ฆฌ๋ชจํ…”
GIST_Signboard/text_images/0/gist_13_4.jpg ๋ด‰์ฅฌ๋ฅด๋…ธ๋ž˜ํ™€
GIST_Signboard/text_images/0/gist_13_6.jpg ์œ ํฅ์ฃผ์ 
GIST_Signboard/text_images/0/gist_13_7.jpg ๊น๋А
GIST_Signboard/text_images/0/gist_13_10.jpg ์‹œํฌ๋ฆฟ์œ ํฅ
GIST_Signboard/text_images/0/gist_13_11.jpg ์œ ํฅ์ฃผ์ 
GIST_Signboard/text_images/0/gist_14_1.jpg ๋ด‰์ฅฌ๋ฅด
GIST_Signboard/text_images/0/gist_14_2.jpg ๋…ธ๋ž˜ํ™€
GIST_Signboard/text_images/0/gist_14_4.jpg ์œ ํฅ์ฃผ์ 
GIST_Signboard/text_images/0/gist_15_1.jpg ์ฒด๋ฆฌ๋ชจํ…”
GIST_Signboard/text_images/0/gist_16_1.jpg ๊น๋А
GIST_Signboard/text_images/0/gist_16_4.jpg ์‹œํฌ๋ฆฟ์œ ํฅ
GIST_Signboard/text_images/0/gist_16_5.jpg ์œ ํฅ์ฃผ์ 
GIST_Signboard/text_images/0/gist_17_1.jpg ํ‚ค์กฐ๊ฐœ์‚ผํ•ฉ
GIST_Signboard/text_images/0/gist_17_2.jpg ์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_17_3.jpg ๋งˆํฌ
GIST_Signboard/text_images/0/gist_17_4.jpg ์„ ์žฅ
GIST_Signboard/text_images/0/gist_17_5.jpg ์ƒ๋ฌด์ง์˜์ 
GIST_Signboard/text_images/0/gist_17_6.jpg ์ฒด์ธ์ ๋ฌธ์˜
GIST_Signboard/text_images/0/gist_18_2.jpg ํ•ซ๋ฐ”
GIST_Signboard/text_images/0/gist_19_1.jpg ๋ด‰์ฅฌ๋ฅด๋…ธ๋ž˜ํ™€
GIST_Signboard/text_images/0/gist_19_4.jpg โ–ก๋ฐ”
GIST_Signboard/text_images/0/gist_20_1.jpg ์ฒด๋ฆฌํ˜ธํ…”
GIST_Signboard/text_images/0/gist_20_4.jpg ํ•ซ๋ฐ”
GIST_Signboard/text_images/0/gist_21_1.jpg ๊ฒฝ๋ณต๊ถ
GIST_Signboard/text_images/0/gist_21_2.jpg ์‚ผ๊ณ„ํƒ•
GIST_Signboard/text_images/0/gist_21_3.jpg ์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_22_1.jpg ๊ฒฝ๋ณต๊ถ
GIST_Signboard/text_images/0/gist_22_2.jpg ์‚ผ๊ณ„ํƒ•
GIST_Signboard/text_images/0/gist_22_3.jpg ์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_23_1.jpg ๊น€ํ˜•์ œ
GIST_Signboard/text_images/0/gist_23_2.jpg ๊ณ ๊ธฐ์˜์ฒ ํ•™
GIST_Signboard/text_images/0/gist_23_4.jpg ํ”„๋ฆฌ๋ฏธ์—„ ํ•œ์šฐ
GIST_Signboard/text_images/0/gist_23_5.jpg ์ด๋ฒ ๋ฆฌ์ฝ” ์„ธ๋ฅด๋„
GIST_Signboard/text_images/0/gist_23_6.jpg ํ”„๋ฆฌ๋ฏธ์—„์ˆ™์„ฑ๋ผ์ง€
GIST_Signboard/text_images/0/gist_24_1.jpg ๊น€ํ˜•์ œ
GIST_Signboard/text_images/0/gist_24_2.jpg ๊ณ ๊ธฐ์˜์ฒ ํ•™
GIST_Signboard/text_images/0/gist_24_4.jpg ํž๋ง
GIST_Signboard/text_images/0/gist_24_5.jpg ํƒ€์ด์•„๋กœ๋งˆ
GIST_Signboard/text_images/0/gist_24_7.jpg ํ”„๋ฆฌ๋ฏธ์—„
GIST_Signboard/text_images/0/gist_24_8.jpg ํ•œ์ •์‹
GIST_Signboard/text_images/0/gist_24_9.jpg ํ•œ์‹๋ ˆ์Šคํ† ๋ž‘
GIST_Signboard/text_images/0/gist_24_10.jpg ์ฒœ์†ก
GIST_Signboard/text_images/0/gist_24_11.jpg ์˜ˆ์•ฝ๋ฌธ์˜
GIST_Signboard/text_images/0/gist_25_1.jpg ์ˆœํƒ„
GIST_Signboard/text_images/0/gist_25_2.jpg ๊ณ ๊ธฐ
GIST_Signboard/text_images/0/gist_25_3.jpg ์ •ํ†ต์—ฐํƒ„๊ตฌ์ด
GIST_Signboard/text_images/0/gist_25_4.jpg ์ˆœ์ˆ˜ํ•œ์—ฐํƒ„
GIST_Signboard/text_images/0/gist_26_1.jpg ํž๋ง
GIST_Signboard/text_images/0/gist_26_5.jpg ์ตœ๊ณ ์˜์Œํ–ฅ์‹œ์„ค
GIST_Signboard/text_images/0/gist_26_6.jpg ํ”ผ๋กœ์—ฐ๋‹จ์ฒด๋ชจ์ž„ํ™˜์˜
GIST_Signboard/text_images/0/gist_26_7.jpg ๋Œ€ํ˜•์ฃผ์ฐจ์žฅ์™„๋น„
GIST_Signboard/text_images/0/gist_27_1.jpg ๊ฐœ๋…ํฌ์ฐจ
GIST_Signboard/text_images/0/gist_27_2.jpg ๋ฐ”๋‹ค๋‹ค
GIST_Signboard/text_images/0/gist_28_1.jpg ๋ณด์ด์Šค
GIST_Signboard/text_images/0/gist_28_3.jpg ๋ฃธ์†Œ์ฃผ๋ฐฉ
GIST_Signboard/text_images/0/gist_29_1.jpg ์ œ์ฃผํ‘๋ผ์ง€์ „๋ฌธ์ 
GIST_Signboard/text_images/0/gist_29_2.jpg ์ œ์ฃผ์‚ผ๋‹คํ‘๋ˆ
GIST_Signboard/text_images/0/gist_29_3.jpg ์˜ˆ์•ฝ๋ฌธ์˜
GIST_Signboard/text_images/0/gist_29_5.jpg ๋ผˆ๋Œ€๊ตฐ
GIST_Signboard/text_images/0/gist_29_6.jpg ๊ตญ๋ฐฅ
GIST_Signboard/text_images/0/gist_29_7.jpg ์ƒ๋ฌด๋ณธ์ 
GIST_Signboard/text_images/0/gist_29_10.jpg ๊ทธ ์ด์ƒ์˜ ๋“ ๋“ ํ•จ
GIST_Signboard/text_images/0/gist_30_1.jpg ์•„์ด
GIST_Signboard/text_images/0/gist_30_2.jpg ๋ฆฌ์Šค
GIST_Signboard/text_images/0/gist_30_3.jpg ์—ฐ์Šต์žฅ
GIST_Signboard/text_images/0/gist_30_4.jpg ๋…ธ๋ž˜
GIST_Signboard/text_images/0/gist_30_6.jpg ๋Œ€ํ˜•๋ฃธ์™„๋น„
GIST_Signboard/text_images/0/gist_30_8.jpg ํž๋ง
GIST_Signboard/text_images/0/gist_30_11.jpg ์ตœ๊ณ ์˜์Œํ–ฅ์‹œ์„ค
GIST_Signboard/text_images/0/gist_30_12.jpg ํ”ผ๋กœ์—ฐ๋‹จ์ฒด๋ชจ์ž„ํ™˜์˜
GIST_Signboard/text_images/0/gist_30_13.jpg ๋Œ€ํ˜•์ฃผ์ฐจ์žฅ์™„๋น„
GIST_Signboard/text_images/0/gist_30_15.jpg ๊ฒฝ๋ณต๊ถ
GIST_Signboard/text_images/0/gist_30_16.jpg ์‚ผ๊ณ„ํƒ•
End of preview. Expand in Data Studio

K-STR-Bench

K-STR-Bench is a unified, challenge-aware benchmark suite for Korean scene text recognition (STR).

It combines three complementary components: AIHub-Standardized, a reproducible large-scale STR resource with nearly 900K valid Korean text instances; KAIST-Corrected, a reconstructed and corrected legacy Korean scene-text dataset; and GIST-Signboard, a newly collected real-world signboard dataset covering challenging outdoor conditions.

The three components are organized under a common Korean-only cropped-STR protocol, supporting reproducible training, cross-dataset evaluation, and challenge-wise robustness analysis.

Code and instructions are available in the GitHub repository.

Myounghun Han, Jin-Hyuk Hong. ACCV 2026.

Subsets

Subset Role Scenes Text Korean-only Train Val Test
AIHub_Standardized Train, val, and in-domain test 533,469 1,937,042 897,055 717,669 89,716 89,670
KAIST_Corrected Evaluation only 2,484 7,516 4,008 โ€” โ€” 4,008
GIST_Signboard Evaluation only 5,185 34,687 16,078 โ€” โ€” 16,078

Korean-only counts are the samples used for K-STR-Bench. Train, val, and test for AIHub_Standardized are that Korean-only split. KAIST_Corrected and GIST_Signboard are test only.

GIST_Signboard and KAIST_Corrected also carry challenge tags on the Korean-only test samples. A sample may have more than one tag.

Subset Artistic Curve Sailent Incomplete Multi-oriented Low-visibility
KAIST_Corrected 563 128 446 56 411 282
GIST_Signboard 2,511 351 2,795 711 4,189 1,070

What this repository contains

GIST_Signboard and KAIST_Corrected images and label tables are included.

AIHub images and the original AIHub annotation files are not included. This repository has the K-STR-Bench metadata, splits, and reviewed transcriptions. Rebuild AIHub_Standardized locally from the AIHub dataset ์•ผ์™ธ ์‹ค์ œ ์ดฌ์˜ ํ•œ๊ธ€ ์ด๋ฏธ์ง€.

Use these label files:

Subset File
AIHub_Standardized text.v2.0.csv
KAIST_Corrected text.v2.0.csv
GIST_Signboard text.v1.0.csv

Each text sample is keyed by text_id. A row points back to its scene with scene_id.

Evaluation

Score matched predictions with word accuracy and jaso-level 1-NED. On GIST_Signboard and KAIST_Corrected, report the same two scores for each challenge tag.

The scoring command and file format are documented in the GitHub repository.

License

This benchmark does not use one license for every file.

Resource Terms
GIST_Signboard images and annotations CC BY 4.0
K-STR-Bench metadata, ids, splits, and author-written labels CC BY 4.0
KAIST_Corrected CC BY-SA 3.0
Original AIHub images and annotations Not redistributed. AIHub Terms of Use

CC BY 4.0 covers only material created and released by the K-STR-Bench authors. It does not cover the original AIHub dataset.

Citation

@inproceedings{han2026kstrbench,
    title     = {K-STR-Bench: A Challenge-Aware Benchmark Suite for Korean Scene Text Recognition},
    author    = {Han, Myounghun and Hong, Jin-Hyuk},
    booktitle = {Asian Conference on Computer Vision (ACCV)},
    year      = {2026}
}

Also cite the original AIHub and KAIST datasets when using those subsets.

Downloads last month
56