Self-Supervised Monocular Depth Underwater

This repo is for Self-Supervised Monocular Depth Underwater paper which can be found here: https://arxiv.org/abs/2210.03206

The work is mostly based on DiffNet which can be found here: https://github.com/brandleyzhou/DIFFNet

requirements:

matplotlib==3.4.2

numpy==1.21.2

opencv-python==4.5.2.52

Pillow==8.4.0

scikit-image==0.18.3

scipy==1.7.1

tensorboard==2.7.0

tensorboardX==2.4

torch==1.10.1

torchvision==0.2.1

Running training over all 4 FLC datasets together and evaluating on each one seperatly (ar all together):

#!/bin/bash echo sleeping.. sleep 18000 dname=FLC_4DS_tiny_sky date=20220706 ds=FLC_4DS_tiny_sky ##################################################3

train

echo "train diffnet flc2 pytorch" run_name=${date}FLC${dname}

echo "${run_name}_test" python train.py --png
--model_name=$run_name
--data_path=
--dataset="uc"
--split=${dname}
--height=480
--width=640
--batch_size=8
--num_epochs=20
--load_weights_folder=
--do_flip
--use_corrLoss
--use_lvw
--use_recons_net

evaluate - option 1: run on each sub ds by it self

for ds in uc flatiron tiny do echo "running flc_new evaluation on ${ds}" python evaluate_depth.py
--model_name="${run_name}eval${ds}"
--dataset=uc
--eval_mono
--load_weights_folder=
--data_path
--save_pred_disps
--use_depth
--eval_split=${ds}
--eval_sky done

evaluate - option 2: run on the unified ds:

echo "running flc_new evaluation on ${ds}" python evaluate_depth.py
--model_name="${run_name}eval${ds}"
--dataset=uc
--eval_mono
--load_weights_folder="/home/samitai/Work/myDIFFNet/models/${run_name}/models/weights_last"
--data_path /home/samitai/Work/Datasets/ANSFL/allData2
--save_pred_disps
--use_depth
--eval_split=${ds}
--eval_sky