datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
UAVBench
🎉 To the best of our knowledge, UAVBench is the first vision-language benchmark to assess low-altitude UAV image-level and region-level understanding and reasoning capabilities of MLLMs.
📢 News
This is an ongoing project. We will be working on improving it.
📦 Complete evaluation code coming soon! 🚀
📦 Detailed low-altitude UAV MLLMs model inference tutorial coming soon! 🚀
[05/13/2025] 🚀 We release our UAVBench benchmark and UAVIT-1M instruct tunning data to… See the full description on the dataset page: https://huggingface.co/datasets/ZhanYang-nwpu/UAVBench.UAVIT-1M
🎉 UAVIT-1M stands as the first largest and most comprehensive instruction tuning dataset to enhance low-altitude UAV visual understanding abilities, supporting 11 image-level and region-level tasks.
📢 News
This is an ongoing project. We will be working on improving it.
📦 Complete evaluation code coming soon! 🚀
📦 Detailed low-altitude UAV MLLMs model inference tutorial coming soon! 🚀
[05/13/2025] 🚀 We release our UAVBench benchmark and UAVIT-1M instruct… See the full description on the dataset page: https://huggingface.co/datasets/ZhanYang-nwpu/UAVIT-1M.nwpu-resisc45
NWPU-RESISC45
Train/val split manifests extracted from GeoChat_Instruct
for fine-tuning vision-language models on satellite scene classification.
Classes
45 scene categories (e.g. airplane, baseball diamond, beach, bridge, …)
Splits
Split
Samples
train
~25 200
val
~4 725
Format
Each JSON file is a list of objects with fields:
image (relative path), question, answer (class label).
UAV-SVG