Corpus ID: 207853328

SCL: Towards Accurate Domain Adaptive Object Detection via Gradient Detach Based Stacked Complementary Losses

@article{Shen2019SCLTA,
  title={SCL: Towards Accurate Domain Adaptive Object Detection via Gradient Detach Based Stacked Complementary Losses},
  author={Zhiqiang Shen and Harsh Maheshwari and Weichen Yao and Marios Savvides},
  journal={ArXiv},
  year={2019},
  volume={abs/1911.02559}
}
Unsupervised domain adaptive object detection aims to learn a robust detector in the domain shift circumstance, where the training (source) domain is label-rich with bounding box annotations, while the testing (target) domain is label-agnostic and the feature distributions between training and testing domains are dissimilar or even totally different. In this paper, we propose a gradient detach based stacked complementary losses (SCL) method that uses detection losses as the primary objective… Expand
Foreground-focused domain adaption for object detection
TLDR
This work proposes a novel foreground-focused domain adaptation (FFDA) framework which mines the loss of the domain discriminators to concentrate on the backpropagation of foreground loss and reaches the new state-of-the-art accuracy on adapting Cityscape to Foggy Cityscape dataset. Expand
Class-unbalanced domain adaptation for object detection via dynamic weighting mechanism
TLDR
A novel domain adaptation for object detection to reduce class space discrepancy between domains is proposed that can effectively solve the problem of performance degradation caused by the discrepancy in class space and significantly improve the detection accuracy in each domain. Expand
Domain Adaptive Object Detection via Asymmetric Tri-way Faster-RCNN
TLDR
This work proposes an asymmetric tri-way Faster-RCNN (ATF) for domain adaptive object detection, which has two distinct merits: 1) A ancillary net supervised by source label is deployed to learn anCillary target features and simultaneously preserve the discrimination of source domain, which enhances the structural discrimination of domain alignment. Expand
Decoupled Adaptation for Cross-Domain Object Detection
  • Junguang Jiang, Baixu Chen, Jianmin Wang, Mingsheng Long
  • Computer Science
  • ArXiv
  • 2021
TLDR
D-adapt is proposed, namely Decoupled Adaptation, to decouple the adversarial adaptation and the training of the detector and fills the blank of regression domain adaptation in object detection by introducing a bounding box adaptor. Expand
DA-DETR: Domain Adaptive Detection Transformer by Hybrid Attention
TLDR
DA-DETR greatly simplifies traditional domain adaptation pipelines by eliminating sophisticated routines that involve multiple adversarial learning frameworks with different types of features and demonstrates superior accuracy as compared with highly-optimized state-of-the-art approaches. Expand
Multi-Source Domain Adaptation for Object Detection
TLDR
A unified Faster RCNN based framework, termed Divide-and-Merge Spindle Network (DMSN), is proposed, which can simultaneously enhance domain invariance and preserve discriminative power and develops a novel pseudo subnet learning algorithm to approximate optimal parameters of pseudo target subset by weighted combination of parameters in different source subnets. Expand
Cross-domain Object Detection through Coarse-to-Fine Feature Adaptation
TLDR
A novel coarse-to-fine feature adaptation approach to cross-domain object detection that ensures domain knowledge in foreground regions can be effectively transferred and results are state-of-the-art. Expand
Domain Adaptive YOLO for One-Stage Cross-Domain Detection
TLDR
A novel Domain Adaptive YOLO (DA-YOLO) is proposed to improve cross-domain performance for one-stage detectors and demonstrates significant improvement when tested under different cross- domain scenarios. Expand
Attention-based Domain Adaptation for Single Stage Detectors
TLDR
This work introduces an attention mechanism that lets us identify the important regions on which adaptation should focus and outperforms the state-of-the-art domain adaptation technique even though it was designed specifically for this particular detector. Expand
PIT: Position-Invariant Transform for Cross-FoV Domain Adaptation
TLDR
The Position-Invariant Transform (PIT) is proposed to better align images in different domains and can soundly boost the performance on both cross-domain object detection and segmentation for state-of-the-art techniques. Expand
...
1
2
3
...

References

SHOWING 1-10 OF 35 REFERENCES
Self-Training and Adversarial Background Regularization for Unsupervised Domain Adaptive One-Stage Object Detection
TLDR
A weak self-training (WST) method and adversarial background score regularization (BSR) for domain adaptive one-stage object detection that diminishes the adverse effects of inaccurate pseudo-labels to stabilize the learning procedure and effectively improves the performance of the one- stage object detection in unsupervised domain adaptation setting. Expand
Diversify and Match: A Domain Adaptive Representation Learning Paradigm for Object Detection
TLDR
This work introduces a novel unsupervised domain adaptation approach for object detection that outperforms the state-of-the-art methods by a large margin in terms of mean average precision (mAP) on various datasets. Expand
Domain Adaptive Faster R-CNN for Object Detection in the Wild
TLDR
This work builds on the recent state-of-the-art Faster R-CNN model, and design two domain adaptation components, on image level and instance level, to reduce the domain discrepancy, based on $$-divergence theory. Expand
Multi-Adversarial Faster-RCNN for Unrestricted Object Detection
  • Zhenwei He, Lei Zhang
  • Computer Science
  • 2019 IEEE/CVF International Conference on Computer Vision (ICCV)
  • 2019
TLDR
A multi-adversarial Faster-RCNN (MAF) framework for unrestricted object detection, which inherently addresses domain disparity minimization for domain adaptation in feature representation and improves the domain adaptability. Expand
DSOD: Learning Deeply Supervised Object Detectors from Scratch
TLDR
Deeply Supervised Object Detector (DSOD), a framework that can learn object detectors from scratch following the single-shot detection (SSD) framework, and one of the key findings is that deep supervision, enabled by dense layer-wise connections, plays a critical role in learning a good detector. Expand
Strong-Weak Distribution Alignment for Adaptive Object Detection
TLDR
This work proposes an approach for unsupervised adaptation of object detectors from label-rich to label-poor domains which can significantly reduce annotation costs associated with detection, and designs the strong domain alignment model to only look at local receptive fields of the feature map. Expand
Cross-Domain Weakly-Supervised Object Detection Through Progressive Domain Adaptation
TLDR
This paper proposes a two-step progressive domain adaptation technique by fine-tuning the detector on two types of artificially and automatically generated samples, and achieves an improvement of approximately 5 to 20 percentage points in terms of mean average precision (mAP) compared to the best-performing baselines. Expand
Subspace Alignment Based Domain Adaptation for RCNN Detector
TLDR
This paper shows that subspace alignment based domain adaptation of the state of the art RCNN based object detector can still adapt sub- spaces that are localized to the object by obtaining detections from the RCNN detector trained on source and applied on target. Expand
SPLAT: Semantic Pixel-Level Adaptation Transforms for Detection
TLDR
This work proposes a Semantic Pixel-Level Adaptation Transform (SPLAT) approach to detector adaptation that efficiently generates cross-domain image pairs and introduces a more efficient cycle-free method, which exploits pixel-level semantic labels to condition the training of the transformation network. Expand
Exploring Object Relation in Mean Teacher for Cross-Domain Detection
TLDR
This work presents Mean Teacher with Object Relations (MTOR) that novelly remolds Mean Teacher under the backbone of Faster R-CNN by integrating the object relations into the measure of consistency cost between teacher and student modules. Expand
...
1
2
3
4
...