Goal: Research the Impact of Imbalanced and Balanced Maritime Code Flag Datasets on the Performance of Two-Stage Image Detectors (R-CNN, Faster R-CNN and Mask R-CNN).

Main metrics:

• intersection over union (IoU),
• precision and recall,
• average precision (AP),
• mean average precision (mAP),
• F1 score (trade-off between precision and recall).

1. UML

TODO

2. Project organization

├── documentation       <- UML diagrams
├── balancers           <- Package with balancers and utils
│   ├── __init__.py     <- Package identicator
│   ├── smote.py        <- SMOTE balancer (interpolation)
│   ├── adasyn.py       <- ADASYN balancer (interpolation)
│   ├── augmentation.py <- Augmentation balancer (augmenting images like rotations, etc.)
│   ├── autoencoder.py  <- Autoencoder balancer (learning needed!)
│   ├── dgan.py         <- DGAN balancer (learning needed!)
│   ├── balancer.py     <- General balancer with all balancers (aggregating all of the above)
│   ├── annotations.py  <- Annotations module
│   └── configuration_reader.py  <- Balancer configuration reader
├── maritime-flags-dataset    <- Source and balanced flags (A-Z)
│   ├── ADASYN_balanced_flags <- Balanced flags by using ADASYN balancer
│   ├── SMOTE_balanced_flags  <- Balanced flags by using SMOTE balancer
│   ├── AUGMENTATION_balanced_flags  <- Balanced flags by using Augmentation balancer
│   ├── DGAN_balanced_flags  <- Balanced flags by using DGAN balancer
│   ├── AE_balanced_flags    <- Balanced flags by using Autoencoder balancer
│   ├── combined_flags       <- Combined/test images 
│   ├── two_flags            <- Balanced two flags (A and B) per 1000 images
│   └── imbalanced_flags     <- Source folder with imbalanced flags
├── balance.py <- Balancing dataset by using balancers package (BALANCING)
├── balancer_configuration.json <- Balancer configuration
└── detection.py <- Training and testing image detectors (EVALUATING)

3. Balancing approaches

3.1 Augmentation

3.2 SMOTE

3.3 ADASYN

3.4 Autoencoder

3.5 Deep Convolutional GAN