MoRe-UAV

This repository hosts the official resources for MoRe-UAV: A Large-Scale Benchmark for Motion-Aware Visual Grounding in UAV Videos.

Repository Structure

• Main/: primary release materials and project resources.
• Method/: evaluation scripts and baseline folder structure.

Dataset Download

The full dataset is currently available through Google Drive.

• Google Drive: Download Link

To simplify transfer and downloading, we first provide image-feature encoded files for the full dataset. We recommend this as the primary way to access the release.

We additionally provide zip packages of the videos and images for users who prefer to work from the original visual data.

To help users quickly download and verify the dataset format and quality, we also provide a small-scale dataset:

• Reviewer Subset (Google Drive): Download Link

Dataset Format

The released dataset is organized into train/, val/, and test/ splits.

dataset/
├── train/
│   ├── expression.json
│   ├── case_00000001/
│   │   ├── images/
│   │   │   ├── 000001.jpg
│   │   │   ├── 000002.jpg
│   │   │   └── ...
│   │   └── bboxes.json
│   └── ...
├── val/
│   ├── expression.json
│   └── ...
└── test/
    ├── expression.json
    └── ...

Each split contains one expression.json file for the cases in that split.

Each case folder contains:

• images/ with ordered video frames
• bboxes.json with frame-level target annotations for the clip