add homogeneous 8 and 24 subsets

KI-1-AI-Sec / adversarial-yolo

Fork of EAVISE adversarial-yolo from paper "Fooling automated surveillance cameras: adversarial patches to attack person detection"

MIT License

2 stars 2 forks source link

add homogeneous 8 and 24 subsets #5

Open mkrupczak3 opened 10 months ago

mkrupczak3 commented 10 months ago

This PR adds two new data subsets for train_patch.py, airbus-homogeneous-8 and airbus-homogeneous-24 and data folders and patch training configs for each.

Bumped learning rate from default 0.03 to 0.08 for airbus-homogeneous-8 things up a bit. Batch size for airbus-homogeneous-24 is triple this at 0.24, to reap the benefits of training in parallel.

This is an experiment to address issue #2