LiYunfengLYF/UOSTrack

Sample Imbalance Adjustment and Similar Object Exclusion in Underwater Object Tracking

Our method

Our manuscript.

Underwater images and Open-air sequences Hybrid Training

Prepare Dataset

Download RUOD(https://github.com/dlut-dimt/RUOD)

Download fish dataset1(https://universe.roboflow.com/pafd/fish-clean) and fish dataset2(https://public.roboflow.com/object-detection/fish)

Merge fish dataset1 and fish dataset2. Then you get FishExtend dataset. Tools are in ./tracking/tools/

RUOD and FishExtend should look like:

${PROJECT_ROOT}
 -- data
     -- RUOD
         |-- annotations
         |-- images
     -- FishExtend
         |-- annotations
         |-- images

Modify Path

Go to lib/train/admin/local.py to set datasets dir

Then you can training tracker follow OSTrack paradigm

Test

Prepare Dataset

Download UOT100(https://www.kaggle.com/datasets/landrykezebou/uot100-underwater-object-tracking-dataset)

Download UTB180(https://www.kaggle.com/datasets/bastech/utb180)

Put UOT100 and UTB in ./data. It should look like:

${PROJECT_ROOT}
 -- data
     -- UOT100
         |-- AntiguaTurtle
         |-- ArmyDiver1
         |-- ArmyDiver2
         ...
     -- UTB180
         |-- Video_0001
         |-- Video_01
         |-- Video_0002
         ...

Modify Path

Go to lib/test/evaluation/local.py to set datasets dir

Model

Checkpoints will be found here.

Put them to ./output/checkpoints/train/ostrack

using MBPP

Go to lib/test/tracker/ostrack.py. Then set use MDPP is True

     # using kalman filter to head        
     # TODO 
     self.use_kf = False  # True

using Underwater Image Enhancement method

Download UIE model Merge it with external/uie Go to lib/test/tracker/ostrack.py. Then set use_uie is True

     # using kalman filter to head        
     # TODO 
     self.use_uie = False  # True
     self.uie = build_fuinegan() # RGHSUWE, UCM, build_shallowuwnet(), build_ushape()

Evaluation

Raw results can be found here.

• UOT100

Put the UOT100 raw results on $PROJECT_ROOT$/output/test/tracking_results/

python tracking/analysis_results.py # need to modify tracker configs and names

• UTB180

Put the UTB1180 raw results on $PROJECT_ROOT$/output/test/tracking_results/

python tracking/analysis_results.py # need to modify tracker configs and names

Acknowledgments

• Thanks for the OSTrack library, which helps us to quickly implement our ideas.