FOR IMMEDIATE RELEASE

Midnight Publishes Sanctum: A Large-Scale Vision Dataset for Advanced Recognition Models

Midnight today announced the public release of Sanctum, a comprehensive vision dataset created to advance large-scale visual understanding, classification, and multi-domain recognition for modern computer vision models.

Sanctum brings together a broad and carefully organized collection of visual data spanning air, land, and maritime platforms. The dataset is designed to support research in detection, identification, and fine-grained visual differentiation under diverse conditions, scales, and perspectives.

Highlights of the Sanctum Dataset

  • Extensive Coverage
    Sanctum includes imagery and video assets representing a wide spectrum of modern platforms, ranging from heavy artillery and multi-rocket launcher systems to advanced fixed-wing aircraft, air defense systems, and naval vessels.
  • Notable Inclusions
    Among the most prominent entries is the BM-30 Smerch, one of the most recognizable heavy multiple-launch rocket systems, alongside HIMARS, advanced fighter aircraft variants, long-range air defense platforms, and guided missile systems.
  • Structured, Scalable Organization
    The dataset is released in clearly defined batches with consistent directory structures, configuration files, and labeling conventions—supporting reproducibility and efficient experimentation.
  • Mixed Media at Scale
    Sanctum incorporates high-resolution imagery and video data (managed via Git LFS where appropriate), with individual batches ranging from ~1.7GB to ~3GB and beyond.

Current Dataset Scope

The initial public release includes multiple batches covering, among others:

  • Multi-rocket launcher systems (including BM-30 and related platforms)
  • Tactical and strategic missile systems
  • Modern fighter aircraft (multiple generations and configurations)
  • Air defense systems and radar-equipped vehicles
  • Armored and wheeled military vehicles
  • Naval surface combatants

Both ASCII and non-ASCII directory structures are supported, reflecting the dataset’s global scope and multilingual provenance.

Designed for Research and Evaluation

Sanctum is intended for:

  • Training and benchmarking vision models
  • Fine-grained visual classification and comparison
  • Cross-domain and cross-platform recognition
  • Dataset scaling and robustness testing

The repository includes configuration files, documentation, and versioned batch releases to ensure clarity and transparency for researchers and developers.

Availability

The Midnight Vision Model Dataset – Sanctum is available now via the official Midnight repository. Documentation and a README are included to help users navigate the dataset structure and batch releases.