Self-Learning Framework for Visual Recognition
A general self-learning frame-work modified from R-FCN is effective on semi-supervised object de-tection. The proposed framework proposes to train object detectors by faithfully recognizing high-confidence object proposals in a self-paced way, and discovering low-confidence ones for user annotation in a active learning way.
Citing R-FCN @article{dai16rfcn,
Author = {Jifeng Dai, Yi Li, Kaiming He, Jian Sun},
Title = {{R-FCN}: Object Detection via Region-based Fully Convolutional Networks},
Journal = {arXiv preprint arXiv:1605.06409},
Year = {2016}
}