Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Job has been terminated due to a temporary spike in resource usage and may be restarted later.
Error code:   JobManagerCrashedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

MPS-CLIP Data (RSICD & RSITMD)

This repository hosts the dataset packages used in MPS-CLIP, including the RSICD and RSITMD remote sensing image–text datasets.

News

  • 2026-11-08: Added region subimage packages rsicd_subimage.zip and rsitmd_subimage.zip. Each package contains train_sub1/ and test_sub1/ with one folder per original image. Each folder holds one region image per noun mentioned in the captions (<noun>_blackbg.png, full image size, black background outside the region). Unzip them into the matching dataset folder (see Download & Extraction).

Contents

File Size Description
rsicd.zip 485 MB RSICD original images (train/, val/, test/)
rsitmd.zip 576 MB RSITMD original images (train/, val/, test/)
rsicd_subimage.zip 4.11 GB RSICD region subimages (train_sub1/, test_sub1/), 49,212 PNGs
rsitmd_subimage.zip 2.56 GB RSITMD region subimages (train_sub1/, test_sub1/), 26,926 PNGs

Directory Structure

Repository files:

MPS-CLIP_Data/
├── README.md
├── rsicd.zip             # RSICD original images
├── rsitmd.zip            # RSITMD original images
├── rsicd_subimage.zip    # RSICD region subimages
└── rsitmd_subimage.zip   # RSITMD region subimages

After extraction:

rsicd/
├── train/                      # 7,862 training images, e.g. 00001.jpg, viaduct_96.jpg
├── val/                        # 1,966 validation images
├── test/                       # 1,093 test images
├── train_sub1/                 # region subimages for train/ (from rsicd_subimage.zip)
│   ├── 00001/                  # one folder per image, named after the image file without extension
│   │   ├── buildings_blackbg.png   # region of the noun "buildings"
│   │   ├── trees_blackbg.png       # region of the noun "trees"
│   │   └── ...
│   └── ...
└── test_sub1/                  # region subimages for test/ (from rsicd_subimage.zip)
    └── <image_name>/<noun>_blackbg.png

rsitmd/
├── train/                      # 3,435 training images, e.g. airport_4663.tif
├── val/                        # 856 validation images
├── test/                       # 452 test images
├── train_sub1/                 # region subimages for train/ (from rsitmd_subimage.zip)
│   └── <image_name>/<noun>_blackbg.png
└── test_sub1/                  # region subimages for test/ (from rsitmd_subimage.zip)
    └── <image_name>/<noun>_blackbg.png

How the subimages are used: for an annotation entry with "image": "train/port_167.jpg" and "nouns": ["port", "buildings", "trees"], the data loader reads the original image train/port_167.jpg and the region images train_sub1/port_167/port_blackbg.png, train_sub1/port_167/buildings_blackbg.png and train_sub1/port_167/trees_blackbg.png. Nouns without a matching file are skipped. In the training config, point image_root at rsicd/ (or rsitmd/), train_sub_root at train_sub1/, and val_sub_root / test_sub_root at test_sub1/.

Intended Use

These datasets are provided here to support training and evaluation for MPS-CLIP on remote sensing vision–language tasks (e.g., image–text retrieval and/or captioning, depending on the experimental setup).

Download & Extraction

hf download lcrucial1f/MPS-CLIP_Data --repo-type dataset --local-dir .

# Original images (each zip already contains its top-level rsicd/ or rsitmd/ folder)
unzip rsicd.zip
unzip rsitmd.zip

# Region subimages (extract into the matching dataset folder)
unzip rsicd_subimage.zip -d rsicd
unzip rsitmd_subimage.zip -d rsitmd
Downloads last month
262

Models trained or fine-tuned on lcrucial1f/MPS-CLIP_Data