domars16k

A Mars image classification dataset for planetary science research.

Dataset Metadata

• License: CC-BY-4.0 (Creative Commons Attribution 4.0 International)
• Version: 1.0
• Date Published: 2025-05-12
• Cite As: TBD

Classes

This dataset contains the following classes:

• 0: aec
• 1: ael
• 2: cli
• 3: cra
• 4: fse
• 5: fsf
• 6: fsg
• 7: fss
• 8: mix
• 9: rid
• 10: rou
• 11: sfe
• 12: sfx
• 13: smo
• 14: tex

Statistics

• train: 1639 images
• test: 234 images
• val: 469 images
• few_shot_train_2_shot: 30 images
• few_shot_train_1_shot: 2 images
• few_shot_train_10_shot: 150 images
• few_shot_train_5_shot: 75 images
• few_shot_train_15_shot: 225 images
• few_shot_train_20_shot: 300 images
• partition_train_0.01x_partition: 15 images
• partition_train_0.02x_partition: 30 images
• partition_train_0.50x_partition: 753 images
• partition_train_0.20x_partition: 301 images
• partition_train_0.05x_partition: 75 images
• partition_train_0.10x_partition: 151 images
• partition_train_0.25x_partition: 377 images

Few-shot Splits

This dataset includes the following few-shot training splits:

• few_shot_train_2_shot: 30 images
• few_shot_train_1_shot: 2 images
• few_shot_train_10_shot: 150 images
• few_shot_train_5_shot: 75 images
• few_shot_train_15_shot: 225 images
• few_shot_train_20_shot: 300 images

Few-shot configurations:

• 2_shot.csv
• 1_shot.json
• 1_shot.csv
• 10_shot.csv
• 5_shot.csv
• 15_shot.csv
• 10_shot.json
• 20_shot.csv
• 2_shot.json
• 15_shot.json
• 20_shot.json
• 5_shot.json

Partition Splits

This dataset includes the following training data partitions:

• partition_train_0.01x_partition: 15 images
• partition_train_0.02x_partition: 30 images
• partition_train_0.50x_partition: 753 images
• partition_train_0.20x_partition: 301 images
• partition_train_0.05x_partition: 75 images
• partition_train_0.10x_partition: 151 images
• partition_train_0.25x_partition: 377 images

Usage

from datasets import load_dataset

dataset = load_dataset("gremlin97/domars16k")

Format

Each example in the dataset has the following format:

{
  'image': Image(...),  # PIL image
  'label': int,         # Class label
}