MINIMA: Modality‑Invariant Image Matching

Dataset Overview

MINIMA is a unified framework for multimodal image matching, aiming to address challenges in cross‑view and cross‑modality matching. The framework enhances generalization via data augmentation and introduces a simple yet effective data engine that generates a large‑scale dataset containing multiple modalities, diverse scenes, and precise matching labels.

Dataset Details

• Dataset Name: MegaDepth‑Syn Dataset
• Generation Method: Produced from the MegaDepth dataset using the MINIMA data engine, adding six extra modalities: infrared, depth, event, normal, sketch, and painting.
• Release: Available on OpenXLab.

Dataset Download

You can download the dataset with the following commands:

pip install openxlab --no-dependencies
openxlab login
openxlab dataset info --dataset-repo lsxi7/MINIMA
openxlab dataset get --dataset-repo lsxi7/MINIMA
openxlab dataset download --dataset-repo lsxi7/MINIMA --source-path /README.md --target-path /path/to/local/folder

Model Weights Download

• Weight files: minima_lightglue, minima_loftr, minima_roma
• Links: Google Drive or GitHub

Test Datasets

• MegaDepth‑1500‑Syn: Download from megadepth‑1500 and organize.
• RGB‑Infrared Test Dataset: From XoFTR, download via Google Drive.
• MMIM Test Dataset: From Multi‑modality‑image‑matching‑database‑metrics‑methods.
• RGB‑Depth Test Dataset: From DIODE, download via AWS or Baidu Cloud.
• RGB‑Event Test Dataset: From DSEC, download via Google Drive.

Dataset Structure

A recommended folder layout is:

data/
├── METU-VisTIR/
│   ├── index/
│   └── ...
├── Multi-modality-image-matching-database-metrics-methods/
│   ├── Multimodal_Image_Matching_Datasets/
│   └── ...
├── megadepth/
│   └── train/[modality]/Undistorted_SfM/
└── DIODE/
    └── val/
└── DSEC/
    ├── vent_list.txt
    ├── thun_01_a/
    └── ...

Citation

If you use this dataset, please cite:

@article{Jiang2024minima,
  title={MINIMA: Modality Invariant Image Matching},
  author={Jiang, Xingyu and Ren, Jiangwei and Li, Zizhuo and Zhou, Xin and Liang, Dingkang and Bai, Xiang},
  journal={arXiv preprint},
  year={2024},
}