ACCENTECXAI/national-blue-superintelligence

---

language: - en license: cc-by-4.0 task_categories: - object-detection task_ids: - object-detection pretty_name: ACCENTECX AI NATIONAL BLUE SUPERINTELLIGENCE (v1.0) size_categories: - 10K<n<100K tags: - marine-ai - computer-vision - object-detection - coral-reef - fisheries - aquaculture - maritime - underwater

🌊 ACCENTECX AI NATIONAL BLUE SUPERINTELLIGENCE (v1.0)

Philippine marine imagery + bounding boxes for ecology, fisheries, aquaculture, maritime awareness, and underwater robotics.

Overview • Quick Specs • Classes • Structure • Annotation • Uses • Limitations • Roadmap • Citation

---

🧭 Overview

The ACCENTECX AI NATIONAL BLUE SUPERINTELLIGENCE (v1.0) is a curated collection of Philippine marine and coastal imagery annotated for object detection using bounding boxes.

It is designed to support:

• 🐟 Fish detection & abundance estimation (proxy)
• 🪸 Coral condition monitoring (healthy vs. bleached)
• 🗑️ Marine litter detection
• 🛥️ Vessel detection & maritime awareness
• 🧰 Aquaculture infrastructure monitoring
• 🤿 Underwater robotics / ROV perception training

Status: Under structured development
Planned release: May 2026
Formats: COCO (primary), YOLO (export supported)

---

⚡ Quick Specs

Category	Details
Task	Object Detection (Bounding Boxes)
Target Size	8,000–12,000 images
Est. Annotations	~40,000–60,000 boxes (estimated)
Habitats	Reef • Seagrass • Benthic • Aquaculture • Coastal Surface
License	CC BY 4.0
Tools	CVAT (annotation)

---

🏷 Classes

Detection Classes (v1.0)

ID	Class
0	fish
1	fish_school
2	coral_healthy
3	coral_bleached
4	macro_invertebrate
5	aquaculture_structure
6	vessel
7	debris
8	diver

Design choice (v1.0): grouped ecological classes for stable detection performance.
Planned: increased biological granularity in future versions.

---

🗂 Structure

images/
  train/
  val/
  test/

annotations/
  coco_annotations.json
  yolo_labels/          (optional export)

metadata/
  metadata.csv

Split Strategy

• 70% train / 20% validation / 10% test
• Splits planned by collection session (not purely random by frame) to reduce near-duplicate leakage.

---

🌍 Metadata

Each image may include:

• image_file
• location
• site
• habitat (reef / seagrass / benthic / aquaculture / coastal_surface)
• depth_min, depth_max (meters, if available)
• visibility (clear / moderate / turbid, if available)
• date (if available)
• source_dive (collection session identifier)

---

🛠 Annotation Protocol

Click to expand annotation rules (CVAT)

Tool & Type

• Tool: CVAT
• Type: Bounding boxes (rectangles)

Core Rules

• Minimum size: skip objects that are too small / unidentifiable
• Dense schools: label as fish_school using one grouped box
• Overlap: annotate separately if visually distinguishable
• Occlusion: skip heavily occluded objects unless clearly identifiable
• QA: reviewer validation + consistency checks

---

🎯 Uses

Direct Use

• Marine object detection benchmarking
• Fish detection + approximate abundance estimation
• Coral condition monitoring
• Marine litter detection systems
• Aquaculture infrastructure monitoring automation
• ROV/AUV perception training

Out-of-Scope Use

• Full species taxonomy modeling (v1.0 uses grouped classes)
• Precision biomass estimation without calibration/scale instrumentation
• Identification of individuals (human recognition/biometrics)
• Safety-critical autonomous navigation without additional validation

---

⚠ Limitations

• Geographic/site distribution may be uneven depending on contributions
• Water visibility and lighting variability may affect generalization
• Some classes may be underrepresented (e.g., debris, vessels)
• Grouped taxonomy reduces biological specificity in v1.0
• Bounding boxes are coarser than segmentation for coverage/area estimation

---

🔮 Roadmap

v1.0

• Core ecological + human-activity detection classes
• COCO primary export, YOLO optional export
• Habitat metadata (reef/seagrass/benthic/aquaculture/coastal_surface)

v2.0 (Planned)

• Expanded invertebrate subclasses (data-driven)
• Fish subgrouping (e.g., reef_fish vs pelagic_fish)
• Optional instance segmentation for coral/seagrass
• Expanded regional coverage and benchmarks

---

📚 Citation

@dataset{accentecx_open_marine_v1,
  author    = {ACCENTECX AI},
  title     = {ACCENTECX Open Marine Dataset (v1.0)},
  year      = {2026},
  publisher = {Hugging Face}
}

---

🏢 About ACCENTECX AI

ACCENTECX AI develops marine-focused artificial intelligence systems and structured datasets supporting fisheries monitoring, coral reef analysis, aquaculture analytics, maritime safety, and autonomous marine systems.

Website: https://accentecxai.com