AquaProbe: Efficient Adaptation of Frozen Vision Models for Underwater Images
AquaProbe: Efficient Adaptation of Frozen Vision Models for Underwater Images
This paper introduces AquaProbe, a method for adapting frozen vision models to underwater imagery by measuring domain gaps based on water type and imaging range. Using paired clean and physically rendered images, the approach allocates a fixed LoRA budget according to the measured divergence profile. The proportional LoRA rank allocation achieves a segmentation mean Intersection-over-Union (mIoU) of 70.1, outperforming uniform allocation by 1.9 points with the same parameter budget of 0.59 million adapter parameters. A three-way router conditioned on estimated condition clusters further achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
The study proposes AquaProbe, a method that measures domain gaps in underwater vision to allocate LoRA adapters non-uniformly, achieving a segmentation mIoU of 70.1 with 0.59 million adapter parameters on paired clean and physically rendered images.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
mIoU
70.1
View original evidence
ion-over-Union (mIoU) for the proportional LoRA rank allocation, based on the measured profile, is 70.1; this is a better result than uniform LoRA rank allocation by 1.9 mIoU points while having the same parameter budget of 0.59 million adapter parameters. The three-way router, conditioned on estimated condition clusters, achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
Open source S4mIoU improvement
1.9
View original evidence
ion-over-Union (mIoU) for the proportional LoRA rank allocation, based on the measured profile, is 70.1; this is a better result than uniform LoRA rank allocation by 1.9 mIoU points while having the same parameter budget of 0.59 million adapter parameters. The three-way router, conditioned on estimated condition clusters, achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
Open source S4mIoU on SUIM split
70.6
View original evidence
ion-over-Union (mIoU) for the proportional LoRA rank allocation, based on the measured profile, is 70.1; this is a better result than uniform LoRA rank allocation by 1.9 mIoU points while having the same parameter budget of 0.59 million adapter parameters. The three-way router, conditioned on estimated condition clusters, achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
Open source S4AUDRC score
63.4
View original evidence
ion-over-Union (mIoU) for the proportional LoRA rank allocation, based on the measured profile, is 70.1; this is a better result than uniform LoRA rank allocation by 1.9 mIoU points while having the same parameter budget of 0.59 million adapter parameters. The three-way router, conditioned on estimated condition clusters, achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ion-over-Union (mIoU) for the proportional LoRA rank allocation, based on the measured profile, is 70.1; this is a better result than uniform LoRA rank allocation by 1.9 mIoU points while having the same parameter budget of 0.59 million adapter parameters. The three-way router, conditioned on estimated condition clusters, achieves 70.6 mIoU on the standard SUIM split and 63.4 AUDRC under held-out degradation.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org