Abstract:We present a novel approach to perform the unsupervised domain adaptation for object detection through forward-backward cyclic (FBC) training. Recent adversarial training based domain adaptation methods have shown their effectiveness on minimizing domain discrepancy via marginal feature distributions alignment. However, aligning the marginal feature distributions does not guarantee the alignment of class conditional distributions. This limitation is more evident when adapting object detectors as the domain discrepancy is larger compared to the image classification task, e.g. various number of objects exist in one image and the majority of content in an image is the background. This motivates us to learn domain invariance for category level semantics via gradient alignment. Intuitively, if the gradients of two domains point in similar directions, then the learning of one domain can improve that of another domain. To achieve gradient alignment, we propose Forward-Backward Cyclic Adaptation, which iteratively computes adaptation from source to target via backward hopping and from target to source via forward passing. In addition, we align low-level features for adapting holistic color/texture via adversarial training. However, the detector performs well on both domains is not ideal for target domain. As such, in each cycle, domain diversity is enforced by maximum entropy regularization on the source domain to penalize confident source-specific learning and minimum entropy regularization on target domain to intrigue target-specific learning. Theoretical analysis of the training process is provided, and extensive experiments on challenging cross-domain object detection datasets have shown the superiority of our approach over the state-of-the-art.

Unsupervised Domain Adaptation for Object Detection Through Mixed-Domain and Co-Training Learning

Multi-View Domain Adaptive Object Detection on Camera Networks.

Joint Feature-Level And Pixel-Level Domain Adaption For Object Detection In The Wild

Object Detection Based on Semi-Supervised Domain Adaptation for Imbalanced Domain Resources.

Spatial Alignment for Unsupervised Domain Adaptive Single-Stage Object Detection

Multi-Level Domain Adaptive Learning For Cross-Domain Detection

Multi-source unsupervised domain adaptation for object detection

Self-Training-Based Unsupervised Domain Adaptation for Object Detection in Remote Sensing Imagery

Domain Adaption Object Detection with Global - Local Contrastive Learning and Co-training Network.

Unsupervised Domain Adaptation for Object Detection via Cross-Domain Semi-Supervised Learning

Unsupervised Domain Adaptive Object Detection using Forward-Backward Cyclic Adaptation

Multilevel Unsupervised Domain Adaptation for Single-Stage Object Detection in Remote Sensing Images

Progressive Domain Adaptation for Object Detection

Multi-Target Domain Adaptation via Unsupervised Domain Classification for Weather Invariant Object Detection

Cross-domain Object Detection through Coarse-to-Fine Feature Adaptation

Uncertainty-Aware Unsupervised Domain Adaptation in Object Detection

Unsupervised Mixed Multi-Target Domain Adaptation for Remote Sensing Images Classification

Class-unbalanced Domain Adaptation for Object Detection Via Dynamic Weighting Mechanism

Domain Adaptation using Self-Training with Mixup for One-Stage Object Detection

Learning a Domain Classifier Bank for Unsupervised Adaptive Object Detection

Unsupervised Domain Adaptation for Monocular 3D Object Detection Via Self-training