1책임연구원, 한국항공우주연구원 국가위성정보활용지원센터, 대전광역시 유성구 과학로 169-84, 34133, 대한민국
2교수, 과학기술연합대학원대학교 항공우주시스템공학전공, 대전광역시 유성구 과학로 169-84, 34133, 대한민국
3석사과정생, 과학기술연합대학원대학교 항공우주시스템공학전공, 대전광역시 유성구 과학로 169-84, 34133, 대한민국
1Principal Researcher, National Satellite Operation & Application Center, Korea Aerospace Research Institute (KARI), 169-84 Gwahak-ro, Yuseong-gu, 34133 Daejeon, South Korea
2Professor, Major in Aerospace System Engineering, University of Science and Technology (UST), 169-84 Gwahak-ro, Yuseong-gu, 34133 Daejeon, South Korea
3Master Student, Major in Aerospace System Engineering, University of Science and Technology (UST), 169-84 Gwahak-ro, Yuseong-gu, 34133 Daejeon, South Korea
Copyright © 2025 GeoAI Data Society
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Conflict of Interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Funding Information
This work was supported by the Satellite Data Applications (No. FR25J00) project through the Korea Aerospace Research Institute (KARI).
Data Availability Statement
The data that support the findings of this study are openly available in DataON at https://doi.org/10.22711/idr/1083.
MB, motorboat; SB, sailboat; TB, tugboat; BG, barge; FB, fishing boat; FR, ferry; CS, cargo ship; OT, oil tanker; DS, drillship; WS, warship; FT, fighter jet; LM, large military aircraft; SC, small civilian aircraft; LC, large civilian aircraft; HC, helicopter; SV, small vehicle; TR, truck; BS, bus; TN, train; IC, individual container; GC, group container; CR, crane; BR, bridge; DM, dam; ST, storage tank; SF, sports field; SD, stadium; SP, swimming pool; RA, roundabout; HP, helipad; WG, wind generator; AF, aquaculture facility; OR, ocean research facility.
Question type | Proportion (%) |
---|---|
Detailed description | 60 |
Object counting | 10 |
Bounding box-based object detection capability | 10 |
Complex reasoning | 20 |
Model | Accuracy (%) |
---|---|
LLaVA-HR (7B) baseline | 54.02 |
LLaVA-HR (7B) fine-tuned | 78.57 |
LLaVA-HR-X (13B) baseline | 72.27 |
LLaVA-HR-X (13B) fine-tuned | 86.59 |
GPT-4o | 82.43 |
Essential |
||
---|---|---|
Field | Sub-Category | |
Title of Dataset | KOMPSAT-3/3A Image-Text Dataset for Training Large Multimodal Dataset | |
DOI | https://doi.org/10.22711/idr/1083 | |
Category | Geoscientific Information | |
Temporal Coverage | 2024.01.-2024.12. | |
Spatial Coverage | Address | Worldwide |
WGS84 Coordinates | ||
Personnel | Name | Han Oh |
Affiliation | Korea Aerospace Research Institute | |
ohhan@kari.re.kr | ||
CC License | CC BY-NC | |
Optional |
||
Field | Sub-Category | |
Summary of Dataset | KOMPSAT-3/3A Image-Text Dataset for Training Large Multimodal Models | |
Project | ||
Instrument |
Class | MB | SB | TB | BG | FB | FR | CS | OT | DS | WS | FT | LM | SC | LC | HC | SV | TR |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Train | 31,469 | 5,536 | 409 | 1,218 | 4,231 | 1,678 | 768 | 195 | 55 | 320 | 827 | 325 | 820 | 1,265 | 605 | 501,394 | 42,776 |
Val | 3,105 | 967 | 70 | 198 | 538 | 195 | 184 | 17 | 17 | 66 | 95 | 17 | 80 | 170 | 81 | 70,055 | 6,153 |
Test | 5,296 | 594 | 104 | 239 | 625 | 209 | 109 | 40 | 17 | 23 | 198 | 16 | 159 | 258 | 197 | 69,617 | 6,441 |
Total | 39,870 | 7,097 | 583 | 1,655 | 5,394 | 2,082 | 1,061 | 252 | 89 | 409 | 1,120 | 358 | 1,059 | 1,693 | 883 | 641,066 | 55,370 |
Class | BS | TN | IC | GC | CR | BR | DM | ST | SF | SD | SP | RA | HP | WG | AF | OR | Total |
Train | 11,133 | 17,332 | 24,005 | 18,362 | 1,754 | 497 | 262 | 5,486 | 2,049 | 118 | 7,982 | 842 | 989 | 181 | 1,618 | 11 | 686,512 |
Val | 1,356 | 3,712 | 3,781 | 3,292 | 283 | 79 | 47 | 1,041 | 325 | 20 | 1,269 | 146 | 114 | 22 | 144 | 2 | 97,641 |
Test | 1,200 | 1,950 | 4,481 | 3,294 | 296 | 83 | 26 | 606 | 370 | 20 | 1,100 | 155 | 165 | 16 | 357 | 3 | 98,264 |
Total | 13,689 | 22,994 | 32,267 | 24,948 | 2,333 | 659 | 335 | 7,133 | 2,744 | 158 | 10,351 | 1,143 | 1,268 | 219 | 2,119 | 16 | 882,417 |
Question type | Proportion (%) |
---|---|
Detailed description | 60 |
Object counting | 10 |
Bounding box-based object detection capability | 10 |
Complex reasoning | 20 |
Model | Accuracy (%) |
---|---|
LLaVA-HR (7B) baseline | 54.02 |
LLaVA-HR (7B) fine-tuned | 78.57 |
LLaVA-HR-X (13B) baseline | 72.27 |
LLaVA-HR-X (13B) fine-tuned | 86.59 |
GPT-4o | 82.43 |
Essential |
||
---|---|---|
Field | Sub-Category | |
Title of Dataset | KOMPSAT-3/3A Image-Text Dataset for Training Large Multimodal Dataset | |
DOI | ||
Category | Geoscientific Information | |
Temporal Coverage | 2024.01.-2024.12. | |
Spatial Coverage | Address | Worldwide |
WGS84 Coordinates | ||
Personnel | Name | Han Oh |
Affiliation | Korea Aerospace Research Institute | |
ohhan@kari.re.kr | ||
CC License | CC BY-NC | |
Optional |
||
Field | Sub-Category | |
Summary of Dataset | KOMPSAT-3/3A Image-Text Dataset for Training Large Multimodal Models | |
Project | ||
Instrument |
MB, motorboat; SB, sailboat; TB, tugboat; BG, barge; FB, fishing boat; FR, ferry; CS, cargo ship; OT, oil tanker; DS, drillship; WS, warship; FT, fighter jet; LM, large military aircraft; SC, small civilian aircraft; LC, large civilian aircraft; HC, helicopter; SV, small vehicle; TR, truck; BS, bus; TN, train; IC, individual container; GC, group container; CR, crane; BR, bridge; DM, dam; ST, storage tank; SF, sports field; SD, stadium; SP, swimming pool; RA, roundabout; HP, helipad; WG, wind generator; AF, aquaculture facility; OR, ocean research facility.