SeafloorGenAI

← Back to all benchmarks

Date: 2024-12-13

Name: SeafloorGenAI

Domain: Climate & Earth Science

Focus: Large-scale vision-language dataset for seafloor mapping and geological classification

Task Types: Image segmentation, Vision-language QA

Metrics: Segmentation pixel accuracy, QA accuracy

Models: SegFormer, ViLT-style multimodal models

AI/ML Motif: Reasoning & Generalization

Resources

Benchmark: Visit

Keywords

sonar imagery vision-language seafloor mapping segmentation QA

Citation

• Kien X. Nguyen, Fengchun Qiao, Arthur Trembanis, and Xi Peng. Seafloorai: a large-scale vision-language dataset for seafloor geological survey. 2024. URL: https://arxiv.org/abs/2411.00172, arXiv:2411.00172.

@misc{nguyen2024seafloor,
  archiveprefix = {arXiv},
  author = {Kien X. Nguyen and Fengchun Qiao and Arthur Trembanis and Xi Peng},
  eprint = {2411.00172},
  primaryclass = {cs.CV},
  title = {SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey},
  url = {https://arxiv.org/abs/2411.00172},
  year=2024
}

Ratings

CategoryRating

Software

3.00

Data processing code is publicly available, but no full benchmark framework or runnable model implementations are provided yet.

Specification

5.00

Tasks (image segmentation and vision-language QA) are clearly defined with geospatial and multimodal objectives well specified.

Dataset

5.00

Large-scale, well-annotated sonar imagery dataset with segmentation masks and natural language descriptions; curated with domain experts.

Metrics

5.00

Standard segmentation pixel accuracy and QA accuracy metrics are clearly specified and appropriate for the tasks.

Reference Solution

4.00

Some baseline models (e.g., SegFormer, ViLT-style) are mentioned, but reproducible code or pretrained weights are not fully available yet.

Documentation

4.00

Dataset description and data processing instructions are provided, but tutorials and benchmark usage guides are limited.

Average rating: 4.33/5

Radar plot

Edit: edit this entry