WaterWave: Underwater Image/Video Enhancement

WaterWave is a BasicSR-based project for underwater enhancement.
It focuses on common underwater degradations such as:

• color cast (blue/green shift)
• low contrast and haze-like appearance
• texture/detail attenuation
• temporal flicker and instability in video sequences

The current primary training config uses the WaveField model in options/train/0.yml.

---

Project Structure (Relevant to This Work)

• options/train/0.yml: main training config (model_type: WaveField)
• basicsr/train.py: training entry
• basicsr/test.py: test entry
• scripts/dist_train.sh: distributed training script
• scripts/dist_test.sh: distributed testing script
• flow_estimate/: optical-flow preprocessing and related tools

---

Environment Setup

Recommended: Python 3.8+ and PyTorch 1.10+ (matching your CUDA version).

cd WaterWave
pip install -r requirements.txt
pip install -v -e .

---

Data Preparation

The default paths in options/train/0.yml include:

• datasets/uw3_mini (GT/training data)
• dataroot_flow (optical flow)
• dataroot_flow_conf (optical-flow confidence)

Update these paths to your local environment before running training.

---

Training

Single-GPU Training

python basicsr/train.py -opt options/train/0.yml

Multi-GPU Training

bash scripts/dist_train.sh 2 options/train/0.yml

---

Testing / Validation

Single-GPU Testing

python basicsr/test.py -opt options/train/0.yml

Multi-GPU Testing

bash scripts/dist_test.sh 2 options/train/0.yml

---

Outputs and Logs

Training/testing outputs are typically written to:

• experiments/<exp_name>/
• results/<exp_name>/

<exp_name> comes from the name field in the config (currently wavewater).

---

Notes

1. basicsr/train.py currently contains a hardcoded path append:

   • sys.path.append('/home/ubuntu/data/code/waterwave_code')
   • If your path is different, switch this to a project-relative path to avoid cross-machine issues.

2. options/train/0.yml is a runnable template. Before formal experiments, tune it based on your data scale:

   • total_iter
   • batch_size_per_gpu
   • num_frame
   • learning rate and scheduler

---

Acknowledgement

This project is built on top of BasicSR.
Please refer to LICENSE and LICENSE.txt for upstream licensing and third-party notices.