My implementations of a variety of algorithms and models for the weakly supervised object detection problem, based on facebookresearch/detectron2
- Python 3
- Latest Pytorch (1.3.1 as of last readme update)
- Latest Torchvision (0.4.2 as of last readme update)
- Detectron2
Below are the models currently implemented in this repository, as well as their performance using VGG16 as a backbone.
- Weakly Supervised Deep Detection Networks (Bilen et. al., 2015) (26.17, no spatial regularisation)
- Multiple Instance Detection Network with Online Instance Classifier Refinement (Tang et. al., 2017) (40.83)
- PCL: Proposal Cluster Learning for Weakly Supervised Object Detection (Tang et. al., 2018) (44.21)