Problem
Research questions and friction points this paper is trying to address.
Lack of challenging datasets for aquatic animal segmentation
Need for effective multi-model fusion in segmentation
Improving training with hard sample synthesis
GUNNEL: Guided Mixup Augmentation and Multi-View Fusion for Aquatic Animal Segmentation
🤖 AI Summary
Existing aquatic animal segmentation research is hindered by the absence of a benchmark dataset exhibiting sufficient species diversity and realistic challenges, coupled with limited model robustness. To address this, we introduce AAS—the first fine-grained segmentation dataset specifically designed for aquatic animals—featuring multi-species annotations and complex real-world conditions including severe illumination variations and occlusions. We further propose GUNNEL, a novel framework that synergistically integrates guided Mixup augmentation (to synthesize hard samples) and multi-view ensemble learning (jointly leveraging feature-level and decision-level fusion) to overcome the generalization limitations of single-model approaches. Extensive experiments on AAS demonstrate that GUNNEL significantly outperforms state-of-the-art methods. Both the AAS dataset and GUNNEL source code are publicly released, establishing a standardized, reproducible benchmark and technical foundation for underwater visual segmentation.
📝 Abstract
Recent years have witnessed great advances in object segmentation research. In addition to generic objects, aquatic animals have attracted research attention. Deep learning-based methods are widely used for aquatic animal segmentation and have achieved promising performance. However, there is a lack of challenging datasets for benchmarking. In this work, we build a new dataset dubbed Aquatic Animal Species. We also devise a novel GUided mixup augmeNtatioN and multi-modEl fusion for aquatic animaL segmentation (GUNNEL) that leverages the advantages of multiple segmentation models to effectively segment aquatic animals and improves the training performance by synthesizing hard samples. Extensive experiments demonstrated the superiority of our proposed framework over existing state-of-the-art instance segmentation methods. The code is available at https://github.com/lmquan2000/mask-mixup. The dataset is available at https://doi.org/10.5281/zenodo.8208877 .
Innovation
Methods, ideas, or system contributions that make the work stand out.
Guided mixup augmentation for hard samples
Multi-model fusion for segmentation
Novel aquatic animal dataset creation
🔎 Similar Papers
No similar papers found.
