TransUTD: Underwater cross-domain collaborative spatial-temporal transformer detector.
Bingxun Zhao1, Xiao Han1, Ruihao Sui1
1School of Mechanical, Electrical & Information Engineering, Shandong University, Weihai, 264209, China.
Summary
Transformer Underwater Spatial-Temporal Cross-domain Collaborative Detection (TransUTD) leverages video sequences to improve underwater object detection. It enhances degraded images using temporal information, achieving state-of-the-art results on benchmark datasets.
Area of Science:
- Computer Vision
- Robotics
- Marine Technology
Background:
- Underwater object detection faces challenges due to degraded image quality.
- Existing methods using single frames or joint enhancement/detection have limitations.
- Temporal information from video sequences offers a promising solution for improved detection.
Purpose of the Study:
- To introduce a novel framework, TransUTD, for robust underwater object detection using spatial-temporal information.
- To reformulate underwater degraded feature representation as a temporal contextual modeling problem.
- To simplify the detection pipeline by eliminating hand-crafted modules.
Main Methods:
- Proposing Transformer Underwater Spatial-Temporal Cross-domain Collaborative Detection (TransUTD).
- Employing a spatial-temporal fusion encoder to aggregate multi-frame features for enhanced semantic representation.
- Utilizing spatial-temporal query interaction for refined localization in complex underwater scenes.
- Implementing a temporal hybrid collaborative decoder for dense supervision via temporal positive queries.
Main Results:
- TransUTD achieves state-of-the-art performance on underwater object detection.
- Demonstrated AP improvements of 1.5% on DUO and 1.9% on the UVID datasets.
- Attained near state-of-the-art performance on ImageNetVID with an AP50 of 86.0%.
Conclusions:
- TransUTD effectively utilizes spatial-temporal information to compensate for feature degradation in underwater imagery.
- The proposed method simplifies the detection pipeline and achieves superior performance.
- The introduction of the UVID dataset facilitates further research in underwater video object detection.
Related Concept Videos
01:11Continuous -time Fourier Transform
957
The Fourier series is instrumental in representing periodic functions, offering a powerful method to decompose such functions into a sum of sinusoids. This technique, however, necessitates modification when applied to nonperiodic functions. Consider a pulse-train waveform consisting of a series of rectangular pulses. When these pulses have a finite period, they can be accurately represented by a Fourier series. Yet, as the period approaches infinity, resulting in a single, isolated pulse, the...
957
01:18Uniform Depth Channel Flow: Problem Solving
543
To calculate the flow rate for a trapezoidal channel, first, identify the bottom width, side slope, and flow depth of the channel. The cross-sectional area (A) corresponding to the depth of flow (y), channel bottom width (B), and side slope (θ) is determined by:Next, calculate the wetted perimeter, which includes the bottom width and the sloped side lengths in contact with the water. Using the values of the cross-sectional area and the wetted perimeter, determine the hydraulic radius by...
543