Repeated multi-angle UAV surveys

Seeing crop fields
in 3D, across seasons.

UAV3DCrop is a large-scale benchmark for 3D reconstruction across crop types, growth stages and repeated UAV acquisitions.

Explore benchmarks ↓ Data & authors GitHub repository ↗
2 nadir · 8 oblique · time ordered
Route 01 / 10 · nadir
UAV image corresponding to the current waypoint
Frame 001 / 100
Full 1,411-point path · 100 sampled images · 90 sec loop
88,830RGB images
4Crop types
3 yearsField seasons
91Benchmark scenes

Field realism,
at research scale.

Four crop systems captured under real outdoor conditions, with substantial variation in canopy structure, illumination and growth stage.

Corn field UAV imageCorn
Soybean field UAV imageSoybean
Wheat field UAV imageWheat
Oat field UAV imageOat

One field.
Eight moments.

Move through the sequence to see how canopy geometry changes over the season. Acquisitions are represented as growth stages to emphasize structural change.

Temporal crop stage 4
Stage 04 / 08
Mid season

Two tracks.
One hard problem.

Evaluate view synthesis and geometry reconstruction under the visual ambiguity of dense, repeating crop canopies.

Track A

Novel view synthesis

Seven scene-optimized methods compared on identical held-out crop regions.

Ground truth and seven novel-view synthesis methods compared across four crops
Same held-out ROI. Rows are corn, soybean, wheat and oat; every column uses the same crop region.
GTNerfactoInstant-NGPSplatfacto Splatfacto-bigMip-SplattingScaffold-GSCityGaussian
Track B

Geometry recovery

Consistent, uncropped views of direct geometry outputs and the four-model benchmark.

MapAnything point-map output
Illustrative MapAnything output · Point map
Benchmarked:MASt3RVGGTPi3MapAnything
Dataset access

Build and evaluate crop-aware 3D reconstruction.

Access the public RGB imagery and depth releases on Hugging Face.

UAV3DCrop ↗ UAV3DCrop Depth ↗
Manuscript

UAV3DCrop: Benchmarking 3D Reconstruction in Repeated Multi-Angle UAV Crop Surveys

Junxiong Zhou, Xuechen Li, Chonghao Qiu, Lang Qiao, Xiaowei Jia, Qi Yang, Chishan Zhang, Leikun Yin, Nanshan You, Vipin Kumar, David Mulla, Ce Yang, Zhenong Jin, and Licheng Liu

Affiliations

University of Minnesota, Twin Cities · University of Wisconsin–Madison · University of Pittsburgh · Max Planck Institute for Biogeochemistry · Boston University · Peking University

Contact

Junxiong Zhou · Licheng Liu

Citation

Dataset citation for now — this entry will be replaced by the formal manuscript citation once it is publicly available.

@misc{uav3dcrop2026,
  title        = {{UAV3DCrop}: Benchmarking {3D} Reconstruction in Repeated Multi-Angle {UAV} Crop Surveys},
  author       = {Zhou, Junxiong and Li, Xuechen and Qiu, Chonghao and Qiao, Lang and Jia, Xiaowei and
                  Yang, Qi and Zhang, Chishan and Yin, Leikun and You, Nanshan and Kumar, Vipin and
                  Mulla, David and Yang, Ce and Jin, Zhenong and Liu, Licheng},
  year         = {2026},
  howpublished = {\url{https://link-dev.github.io/UAV3DCrop/}},
  note         = {Dataset: \url{https://huggingface.co/datasets/Link-Dev/UAV3DCrop}}
}