MMLSv2: A Multimodal Dataset for Martian Landslide Detection in Remote Sensing Imagery

Announcements

• MMLSv2 papers are now available at CVF repository: dataset paper, challenge paper
• MMLSv2 has been accepted at the 4th Workshop on AI for Space (AI4Space) @ CVPR 2026 📣📣📣
• MMLSv2 preprint is available on arXiv
• MMLSv2 is the official dataset for the 1st Mars Landslide Segmentation Challenge (MARS-LS) at the 22nd IEEE/CVFPerception Beyond the Visible Spectrum Workshop @ CVPR 2026.

Summary

We present MMLSv2, a dataset for landslide segmentation on Martian surfaces. MMLSv2 consists of multimodal imagery with seven bands: RGB, digital elevation model, slope, thermal inertia, and grayscale channels. MMLSv2 comprises 664 images distributed across training, validation, and test splits. In addition, an isolated test set of 276 images from a geographically disjoint region from the base dataset is released to evaluate spatial generalization.

Dataset description

Splits and statistics

The distribution of the MMLSv2 dataset across the different splits is summarized below. The foreground ratio (FG) is expressed as the percentage of pixels belonging to landslide regions, including its average (Avg. FG), standard deviation (Std. FG), and minimum–maximum values (Min. FG, Max. FG).

Split	# Images	Avg. FG (%)	Std. FG (%)	Min. FG (%)	Max. FG (%)
Train	465	35.41	25.64	0.02	99.52
Val	66	31.53	24.05	0.08	90.32
Test	133	33.82	25.05	0.10	90.67
Isolated test	276	21.83	17.08	0.01	71.95

Band order

The MMLSv2 dataset consists of seven bands, each representing different spectral or derived information used for analysis. The bands are ordered as follows:

Band	Description
B1	Red
B2	Green
B3	Blue
B4	DEM
B5	Slope
B6	Thermal inertia
B7	Grayscale

Image stats and format

Each sample in the dataset is represented as a multi-channel image with the following characteristics:

• Shape: (128, 128, 7)
• Dtype: float32
• Channels: 7
• Value range: 0.0 to 1.0

Mask stats and format

Each mask in the dataset corresponds to a single-channel annotation map with the following characteristics:

• Shape: (128, 128)
• Dtype: uint8
• Channels: 1 (grayscale)
• Unique values: [0, 1]
• Value range: 0 to 1

Isolated test clarification

Due to the inclusion of the MMLSv2 dataset in the Mars Landslide Segmentation Challenge at PBVS/CVPR, the isolated test set will not be released at this time. It will be made available in the near future.

Usage

from datasets import load_dataset
import numpy as np

# 1. Load the dataset directly from Hugging Face Hub
dataset = load_dataset("MarsLS/MMLSv2")

# 2. Load a sample from the training set
sample = dataset["train"][0]

# 3. Convert to numpy array to preserve all 7 channels from the .tif files
# Shape will be (128, 128, 7)
image_channels = np.array(sample["image"])
mask = np.array(sample["label"])

# 4. Index specific bands based on the dataset structure
rgb_channels = image_channels[:, :, 0:3]       # B1 (Red), B2 (Green), B3 (Blue)
dem_channel = image_channels[:, :, 3]          # B4 (DEM)
slope_channel = image_channels[:, :, 4]        # B5 (Slope)
thermal_inertia = image_channels[:, :, 5]      # B6 (Thermal Inertia)
grayscale = image_channels[:, :, 6]            # B7 (Grayscale)

Citation

If you find this dataset useful, please like ❤️❤️❤️ our repo and cite our papers:

@InProceedings{Paheding_2026_CVPR,
    author    = {Paheding, Sidike and Reyes-Angulo, Abel A. and Ramos, Leo Thomas and Sappa, Angel D. and A, Rajaneesh and B, Hiral P and K.S., Sajin Kumar and Oommen, Thomas},
    title     = {MMLSv2: A Multimodal Dataset for Martian Landslide Detection in Remote Sensing Imagery},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2026},
    pages     = {10329-10338}
}

@InProceedings{Ramos_2026_CVPR,
    author    = {Ramos, Leo Thomas and Reyes-Angulo, Abel and Paheding, Sidike and Sappa, Angel D.},
    title     = {1st Mars Landslide Segmentation Challenge - PBVS 2026},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2026},
    pages     = {7132-7141}
}

Authors and Contact

Sidike Paheding - Fairfield University, USA - spaheding@fairfield.edu

Leo Thomas Ramos - Computer Vision Center, Universitat Autònoma de Barcelona, Spain - ltramos@cvc.uab.cat

Abel Reyes-Angulo - Michigan Technological University, USA - areyesan@mtu.edu

Angel D. Sappa - Computer Vision Center, Universitat Autònoma de Barcelona, Spain - asappa@cvc.uab.cat