Mars-Bench: mb-conequest_seg

This is a binary segmentation dataset focused on identifying volcanic cones on the Martian surface using CTX imagery. It was developed to support global mapping and morphologic analysis of small-scale volcanic landforms. The dataset spans six geographically diverse regions on Mars, capturing substantial variation in cone shape, size, and appearance, making it a challenging benchmark for model generalization. Each sample consists of an image and its corresponding binary mask, with all annotations created and validated by expert geologists to ensure scientific accuracy. Notably, the dataset includes negative samples (images without any cones), which introduces additional complexity by requiring models to correctly predict true negatives rather than detecting cones in every image. We provide metadata indicating which samples are negative, allowing users the flexibility to include or exclude them during training. This information can also be verified using simple image processing techniques.

To explore all Mars-Bench resources, visit our project website:

https://mars-bench.github.io/