Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Introducing VaDA: Novel Image Segmentation Model for Maritime Object Segmentation Using New Dataset

Yongjin Kim, Jinbum Park, Sanha Kang, Hanguen Kim

TL;DR

This work introduces VaDA, a maritime object segmentation model that leverages Vertical and Detail Attention to boost performance in diverse sea conditions. It also proposes IFCP, a holistic real-time evaluation metric, and introduces the OASIs dataset to benchmark segmentation under day, adverse weather, and night scenarios. VaDA achieves state-of-the-art results on OASIs with an IFCP of $0.6422$ and mIoU of $0.7993$, while maintaining practical edge-device feasibility. Together, these contributions advance robust, real-time maritime perception and provide a standardized dataset and evaluation framework for the community.

Abstract

The maritime shipping industry is undergoing rapid evolution driven by advancements in computer vision artificial intelligence (AI). Consequently, research on AI-based object recognition models for maritime transportation is steadily growing, leveraging advancements in sensor technology and computing performance. However, object recognition in maritime environments faces challenges such as light reflection, interference, intense lighting, and various weather conditions. To address these challenges, high-performance deep learning algorithms tailored to maritime imagery and high-quality datasets specialized for maritime scenes are essential. Existing AI recognition models and datasets have limited suitability for composing autonomous navigation systems. Therefore, in this paper, we propose a Vertical and Detail Attention (VaDA) model for maritime object segmentation and a new model evaluation method, the Integrated Figure of Calculation Performance (IFCP), to verify its suitability for the system in real-time. Additionally, we introduce a benchmark maritime dataset, OASIs (Ocean AI Segmentation Initiatives) to standardize model performance evaluation across diverse maritime environments. OASIs dataset and details are available at our website: https://www.navlue.com/dataset

Introducing VaDA: Novel Image Segmentation Model for Maritime Object Segmentation Using New Dataset

TL;DR

This work introduces VaDA, a maritime object segmentation model that leverages Vertical and Detail Attention to boost performance in diverse sea conditions. It also proposes IFCP, a holistic real-time evaluation metric, and introduces the OASIs dataset to benchmark segmentation under day, adverse weather, and night scenarios. VaDA achieves state-of-the-art results on OASIs with an IFCP of and mIoU of , while maintaining practical edge-device feasibility. Together, these contributions advance robust, real-time maritime perception and provide a standardized dataset and evaluation framework for the community.

Abstract

The maritime shipping industry is undergoing rapid evolution driven by advancements in computer vision artificial intelligence (AI). Consequently, research on AI-based object recognition models for maritime transportation is steadily growing, leveraging advancements in sensor technology and computing performance. However, object recognition in maritime environments faces challenges such as light reflection, interference, intense lighting, and various weather conditions. To address these challenges, high-performance deep learning algorithms tailored to maritime imagery and high-quality datasets specialized for maritime scenes are essential. Existing AI recognition models and datasets have limited suitability for composing autonomous navigation systems. Therefore, in this paper, we propose a Vertical and Detail Attention (VaDA) model for maritime object segmentation and a new model evaluation method, the Integrated Figure of Calculation Performance (IFCP), to verify its suitability for the system in real-time. Additionally, we introduce a benchmark maritime dataset, OASIs (Ocean AI Segmentation Initiatives) to standardize model performance evaluation across diverse maritime environments. OASIs dataset and details are available at our website: https://www.navlue.com/dataset
Paper Structure (26 sections, 2 equations, 9 figures, 6 tables)

This paper contains 26 sections, 2 equations, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Maritime Datasets
  4. Semantic Segmentation
  5. Real-Time Segmentation
  6. State-of-the-Art Segmentation
  7. OASIs: Ocean AI Segmentation Initiatives
  8. Dataset Acquisition
  9. Dataset Annotation
  10. Dataset Configuration
  11. Type-1: Day-Time Environment
  12. Type-2: Adverse Weather Environment
  13. Type-3: Night-Time Environment
  14. VaDA: Vertical and Detail Attention
  15. Feature Extraction Backbone
  16. ...and 11 more sections

Figures (9)

  • Figure 1: The comparison of Integrated Figure of Calculation Performance (IFCP) and Frame Per Seconds (FPS) on the Seadronix evaluation dataset, OASIs (Ocean AI Segmentation Initiatives). Red marker represents our proposed Vertical and Detail Attention (VaDA). The test environment matches the inference environment. The experimental results demonstrate that VaDA achieves the best IFCP.
  • Figure 2: The sample image of our measuring Hardware equipment, SxSM200N.
  • Figure 3: The sample images of OASIs (Ocean AI Segmentation Initiatives). (a) shows included daytime scenes (sunny, mild cloudy,back-lit) in OASIs Type-1. (b) includes abnormal weather scenes (rainy, foggy) in Type-2 and (c) includes night-time scenes (dark, dark w/ light source, early evening, and dawn) in Type-3 respectively.
  • Figure 4: The scene distribution of maritime weather conditions in our evaluation dataset, OASIs. Normal weather conditions are on the left side of the distribution bar and Abnormal weather conditions are on the right. "Night" contains night time condition scenes that are difficult for the RGB sensor of the camera to obtain information.
  • Figure 5: The sample images and annotated labels pairs. (Left) shows an original image of OASIs. (Right) shows an annotated label for each input image.
  • ...and 4 more figures