Efficient-USR: Prompt guided Dual-Domain feature information for efficient underwater image super-resolution
Paper Link: https://ieeexplore.ieee.org/document/10889519
This paper has been accepted at the 50th IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2025).
The official repository with Pytorch
In the above diagram the red dotted arrow for up-sampling and blue dotted arrow for down-sampling.
Python 3.9.12
- create a virtual environment
python3 -m venv ./venv_name- activate virtual environment
source venv_name/bin/activate- install dependencies
pip3 install torch torchvision opencv-python matplotlib pyyaml tqdm tensorboardX tensorboard einops thop- Train the Efficient-USR on UFO-120 dataset.
python train.py -v "UFO_2X_32" -p train --train_yaml "train_UFO120_x2_32.yaml"
python train.py -v "UFO_3X_32" -p train --train_yaml "train_UFO120_x3_32.yaml"
python train.py -v "UFO_4X_32" -p train --train_yaml "train_UFO120_x4_32.yaml"- Train the Efficient-USR on USR-248 dataset.
python train.py -v "USR_2X_32" -p train --train_yaml "train_USR248_x2_32.yaml"
python train.py -v "USR_4X_32" -p train --train_yaml "train_USR248_x4_32.yaml"
python train.py -v "USR_8X_32" -p train --train_yaml "train_USR248_x8_32.yaml"python train.py -v "UFO_2X_32" -p finetune --ckpt 277Test the Efficient-USR model using UFO-120 and USR-248 datasets.
| -- | UFO | -- | -- | USR | -- | |
|---|---|---|---|---|---|---|
| Scale | Version | Epoch | Scale | Version | Epoch | |
| 2x | UFO_2X_32 | 277 | 2x | USR_2X_32 | 238 | |
| 3x | UFO_3X_32 | 300 | 3x | USR_4X_32 | 274 | |
| 4x | UFO_4X_32 | 300 | 4x | USR_8X_32 | 292 |
- e.g.,
python test.py -v "UFO_2X_32" --checkpoint_epoch 277 -t tester_Matlab --test_dataset_name "UFO-120"- provide dataset path in env/env.json file
- other configurations are done using yaml files
@inproceedings{pramanick2025efficient,
title={Efficient-USR: Prompt Guided Dual-Domain Feature Information for Efficient Underwater Image Super-Resolution},
author={Pramanick, Alik and Bheda, Utsav and Sur, Arijit},
booktitle={ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages={1--5},
year={2025},
organization={IEEE}
}