Research map / Detection and segmentation / CNN features

LAqua: Laplacian Pyramids for Aquatic Segmentation (Student Abstract)

AAAI 2025 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes LAqua, a lightweight (1.1M-parameter) Laplacian-pyramid segmentation network for marine/aquatic scenes, evaluated on MaSTr1325 and an oil spill drone dataset.

Problem

Semantic segmentation of marine environments for USV autonomous navigation and oil spill detection, where accurate edge perception must be balanced against computational cost.

Approach

A lightweight multi-branch network that uses a three-level Laplacian pyramid decomposition with low, middle, and top residual branches to capture coarse structure and high-frequency edge details, with concatenation-based hierarchical upsampling.

Key results

• On MaSTr1325, LAqua reaches 98.97 mIoU versus 99.91 for BEMSNet and 99.36 for UNet, with 1.1M parameters versus 71.4M for BEMSNet.
• On the Oil Spill dataset, LAqua reaches 94.8 mIoU versus 97.51 for UNet, which has 31M parameters.
• The paper claims up to 28x fewer parameters than other methods while achieving near state-of-the-art accuracy.

Surprising findings

• LAqua is slightly below UNet on the oil spill dataset (94.8 vs 97.51 mIoU) and below the best MaSTr1325 baselines, despite its much smaller size.

Limitations

• Accuracy is below the best baselines on both datasets (e.g., BEMSNet on MaSTr1325, UNet on oil spills).
• Limited experimental detail: no hardware, training data split, or runtime measurements reported.
• Only two datasets evaluated; baseline results for some models are taken on the validation set.

Experimental setup

Datasets: MaSTr1325 Oil Spill Drone dataset

Benchmarks: MaSTr1325 Oil Spill Drone dataset

Metrics: mIoU number of parameters (M)

Baselines: PSPNet DeepLabv3 HRNet UNet WaSRNet BEMSNet

Builds on: Laplacian pyramid WaSRNet BEMSNet

Closest papers by meaning

• SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey (NeurIPS 2024 · similarity 0.84)
• CoralSCOP: Segment any COral Image on this Planet (CVPR 2024 · similarity 0.83)
• Fantastic Animals and Where to Find Them: Segment Any Marine Animal with Dual SAM (CVPR 2024 · similarity 0.83)
• CoralVQA: A Large-Scale Visual Question Answering Dataset for Coral Reef Image Understanding (NeurIPS 2025 · similarity 0.83)
• SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · similarity 0.83)
• GLH-Water: A Large-Scale Dataset for Global Surface Water Detection in Large-Size Very-High-Resolution Satellite Imagery (AAAI 2024 · similarity 0.83)