This is the implementation of our paper Instance-Invariant Domain Adaptive Object Detection via Progressive Disentanglement that has been accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI).
For domain adaptive object detection （DAOD）, it is important to obtain the instance-invariant features that are domain-invariant. To this end, this paper proposes a progressive disentangled framework to disentangle the domain-invariant features from the domain-specific features.