This search addressed “Remote-sensing image object-detection datasets, for remote-sensing detection model training and evaluation”, covering about 3 source platforms with 4 candidates returned; after a per-must-have assessment, 2 meet all must-haves and 2 largely match. The full candidate list and item-by-item basis follow below.
Note: standard search is billed per search, not by whether a match is found; information retrieval does not guarantee a match, and credits are non-refundable on no-match (refunded only if a system fault prevents the report). We provide source links, the search process and per-item verdicts — not raw data; obtain the data yourself under each source’s license.
▸About this sample · delivery boundaryexpand
This page is a sample availability report with the same structure as a real search report; candidates, verdicts and sources are compiled from public data sources with click-through links. Downloadability, field completeness and commercial licensing are subject to each source’s official terms. For item-by-item verified, ready-to-use datasets, our team can deliver a custom build.
The requirement
Framed through the requirements assistant into a base requirement plus must-have / nice-to-have conditions. A match means all must-haves are met; nice-to-haves are a bonus.
Base requirement
Remote-sensing image object-detection datasets, for remote-sensing detection model training and evaluation.
Must-have · verdict baseline
- Oriented (OBB) or horizontal (HBB) box annotations
- Large enough (≥ thousands of images or 100k+ instances)
- Accessible from an official source or authoritative platform
- Clear license terms
Nice-to-have · bonus
This search’s result
A page-level search result; a match means all must-haves are met. For field-by-field verified, ready-to-use datasets, our team can deliver a custom build (quoted separately).
Search coverage: real search across university official sources, dataset repositories and challenge platforms; some sets contain Google Earth imagery and forbid commercial use; community copies need verification against the official release.
Per-must-have verdict
Each candidate first gives its reasoning, then a “Met / TBD / Not met” verdict per must-have with the basis. Match = all your must-haves are met; Near match = on the right track but a condition or two short.
Listed 4 candidate(s), of which 2 match.
DOTA v2.0
1.7M+ objects, 18 classes, oriented boxes (OBB)
Assessment: 1.7M+ annotated objects, 18 classes, oriented boxes (OBB), downloadable from the official page with an official devkit; the license is clearly academic-only.
Note: Contains Google Earth imagery, no commercial use; verify community copies against the official release.
DIOR
20k+ large images, 20 classes, HBB (oriented version too)
Assessment: 20k+ 800×800 images, 20 classes, horizontal boxes (HBB), with an oriented version DIOR-R; downloadable from the official page.
Note: Contains Google Earth imagery; commercial use is risky.
NWPU VHR-10
800 images, 10 classes, HBB, good for getting started
Assessment: 800 images, 10 classes, horizontal boxes (HBB), downloadable from GitHub and free for academic use — annotations and license are fine; but at only 800 images it is clearly below your “≥ thousands of images” threshold, so “large enough” is not met and it suits only getting started / quick validation, hence a near match.
Gaps: Only 800 images — does not meet “≥ thousands of images”; suits getting started or quick validation, not large-scale training.
FAIR1M
1M oriented instances, 37 subclasses, finest taxonomy
Assessment: 40k+ images, 1M oriented-box instances, 5 categories / 37 subclasses, the finest taxonomy; but the full set requires application via the challenge platform and the test set is not public.
Gaps: The full set (with test set) is not public; “full” copies online are usually partial; evaluation must go through the challenge platform.
Turn matches into deep verification
Our team can verify the sources and files item by item, confirm your must-haves are met, and deliver. Email contact@deepsdata.com or add us on WeCom.
Conclusion & next steps
A page-level conclusion; downloadability, field completeness and commercial license are subject to each source’s official terms.
Conclusion
2 candidates (DOTA v2.0, DIOR) meet every must-have item by item: for scale and oriented boxes prefer DOTA; for large horizontal-box images use DIOR. NWPU VHR-10 has only 800 images — below “≥ thousands” — so it is a near match (good for getting started); FAIR1M has the finest taxonomy but its full set needs application and the test set is not public, also a near match.
Next-step advice
Leave your details for our team to run a deep search and download verification, confirming each of your must-haves before delivery.
Team deep-verification
Download verification, field completeness and full matches are confirmed after the team opens and checks them.
Note
DOTA and DIOR contain Google Earth imagery and forbid commercial use; community copies (PaddlePaddle / Hugging Face) need verification against the official release; FAIR1M’s full set and test set are not public, and evaluation goes through the challenge platform. We do not deliver raw data — obtain it yourself under each source’s license.
Want your own availability report?
Tell us your data need and must-have conditions; we assess availability first, then organize the data to be ready to use. Not quite the fit you need? Our team can build a custom dataset — sourced, organized and verified against your must-haves, feasibility assessed first, then you decide.
