Search papers, labs, and topics across Lattice.
The paper introduces a Downstream Task-Inspired Underwater Image Enhancement (DTI-UIE) framework to improve the performance of underwater image recognition tasks by focusing on task-relevant image features. They design a two-branch network with a task-aware attention module, trained using a multi-stage framework and a task-driven perceptual loss. The authors also create a Task-Inspired UIE Dataset (TI-UIED) using task-specific networks to guide the enhancement process.
Forget generic image quality metrics – this underwater image enhancement method boosts downstream task performance by directly optimizing for the features that matter to semantic segmentation and object detection.
In real underwater environments, downstream image recognition tasks such as semantic segmentation and object detection often face challenges posed by problems like blurring and color inconsistencies. Underwater image enhancement (UIE) has emerged as a promising preprocessing approach, aiming to improve the recognizability of targets in underwater images. However, most existing UIE methods mainly focus on enhancing images for human visual perception, frequently failing to reconstruct high-frequency details that are critical for task-specific recognition. To address this issue, we propose a Downstream Task-Inspired Underwater Image Enhancement (DTI-UIE) framework, which leverages human visual perception model to enhance images effectively for underwater vision tasks. Specifically, we design an efficient two-branch network with task-aware attention module for feature mixing. The network benefits from a multi-stage training framework and a task-driven perceptual loss. Additionally, inspired by human perception, we automatically construct a Task-Inspired UIE Dataset (TI-UIED) using various task-specific networks. Experimental results demonstrate that DTI-UIE significantly improves task performance by generating preprocessed images that are beneficial for downstream tasks such as semantic segmentation, object detection, and instance segmentation. The codes are publicly available at https://github.com/oucailab/DTIUIE.
