Harmonizing Transferability and Discriminability for Adapting Object Detectors

By Chaoqi Chen, Zebiao Zheng, Xinghao Ding, Yue Huang, Qi-Yan Dou. Computer Vision and Pattern Recognition, 2020

335 citations recorded by the source corpus.

Recent advances in adaptive object detection have achieved compelling results in virtue of adversarial feature adaptation to mitigate the distributional shifts along the detection pipeline. Whilst adversarial adaptation significantly enhances the transferability of feature representations, the feature discriminability of object detectors remains less investigated. Moreover, transferability and discriminability may come at a contradiction in adversarial adaptation given the complex combinations of objects and the differentiated scene layouts between domains. In this paper, we propose a Hierarchical Transferability Calibration Network (HTCN) that hierarchically (local-region/image/instance) calibrates the transferability of feature representations for harmonizing transferability and discriminability. The proposed model consists of three components: (1) Importance Weighted Adversarial Training with input Interpolation (IWAT-I), which strengthens the global discriminability by re-weighting the interpolated image-level features; (2) Context-aware Instance-Level Alignment (CILA) module, which enhances the local discriminability by capturing the underlying complementary effect between the instance-level feature and the global context information for the instance-level feature alignment; (3) local feature masks that calibrate the local transferability to provide semantic guidance for the following discriminative pattern alignment. Experimental results show that HTCN significantly outperforms the state-of-the-art methods on benchmark datasets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/16c93be6f788711010e0504861e39bd64ea87361

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

A Survey on Negative Transfer (2020, background): “[99] BSP Adversarial TL D, S Common latent space”
Discriminability-Transferability Trade-Off: An Information-Theoretic Perspective (2022, background): “The transferability of deep representations has been studied from various perspectives [23, 36, 11, 43, 52, 24, 3, 7].”
Overcoming Negative Transfer: A Survey (2020, methodology): “ the largest k singular values. Chen et al. [83] also found that learning discriminative features in the shared feature space can signiﬁcantly boost the performance of deep DA approaches. Chen et al. [84] developed a hierarchical transferability calibration network, which hierarchically calibrates t”
Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (2021, background): “However, [4] shows there is an optimality gap between the optimal source hypothesis and the optimal target hypothesis in UDA, if two domains are misaligned at the category-level, we can hardly find a joint hypothesis that simultaneously minimizes the source and target expected errors, leading to a l”
Cross-Domain Adaptive Teacher for Object Detection (2021, methodology): “We compare our method with SCL [35], SWDA [32], DM [19], CRDA [42], HTCN [2], UMT [8], Source (F-RCNN), and Oracle (F-RCNN).”
Unbiased Mean Teacher for Cross-domain Object Detection (2020, methodology): “Many approaches utilize adversarial learning manner with a gradient reverse layer to obtain domain-invariant feature, such as DA-Faster [6], SCDA [56], SWDA [37], SPLAT [47], MAF [17], MDAL [50], CRDA [51], CDN [44], HTCN [4], etc.”