This is the PyTorch implementation of Improving Object Detectors by Exploiting Bounding Boxes for Augmentation Design, Accepted at IEEE ACCESS

Dataset

PASCAL VOC, BDD and Synthetic Fruits datasets were utilized for all our experiments.

Requirements

We tested our code with the following package versions

pytorch 1.12.0
cudatoolkit 11.3.0
Detectron2 0.6

Experiments on Pascal VOC

InterAug

Example command to train a faster-rcnn model with only 10% of training data using InterAug augmentation protocol. python -u interaug_pascalvoc.py --model faster_rcnn --seed 0 --percent 0.1 --augmentation_type InterAug

Similarly for retinanet please run

python -u interaug_pascalvoc.py --model retinanet --seed 0 --percent 0.1 --augmentation_type InterAug

BBAug

Example commands to train using BBAug Augmentation protocol

python -u interaug_pascalvoc.py --model faster_rcnn --seed 0 --percent 0.1 --augmentation_type BBAug

python -u interaug_pascalvoc.py --model retinanet --seed 0 --percent 0.1 --augmentation_type BBAug

ImageAug

Sample commands to train using ImageAug augmentation strategy

python -u imageaug_pascalvoc.py --model faster_rcnn --seed 0 --percent 0.1 --augmentation_type ImageAug

python -u imageaug_pascalvoc.py --model retinanet --seed 0 --percent 0.1 --augmentation_type ImageAug

Experiments on BDD

Sample commands to run on BDD using all three augmentation protocols. Please change the --model argument to desired architecture

python -u interaug_bdd.py --model faster_rcnn --seed 0 --percent 0.1

python -u bbaug_bdd.py --model faster_rcnn --seed 0 --percent 0.1

python -u imageaug_bdd.py --model faster_rcnn --seed 0 --percent 0.1