Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation

Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation. Chen, L., Chen, H., Wei, Z., Jin, X., Tan, X., Jin, Y., & Chen, E. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 7181-7190, 2022.

Paper

Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation [link]

Website abstract bibtex

Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .

@inproceedings{
 title = {Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation},
 type = {inproceedings},
 year = {2022},
 pages = {7181-7190},
 websites = {https://github.com/xiaoachen98/DALN},
 id = {4cac66c7-fa19-38b6-bae1-d3dcc2048340},
 created = {2023-01-06T14:07:13.088Z},
 accessed = {2023-01-06},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-01-13T07:12:08.126Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .},
 bibtype = {inproceedings},
 author = {Chen, Lin and Chen, Huaian and Wei, Zhixiang and Jin, Xin and Tan, Xiao and Jin, Yi and Chen, Enhong},
 booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022}
}

Downloads: 0

{"_id":"9rTgZrYbu74u28dfs","bibbaseid":"chen-chen-wei-jin-tan-jin-chen-reusingthetaskspecificclassifierasadiscriminatordiscriminatorfreeadversarialdomainadaptation-2022","author_short":["Chen, L.","Chen, H.","Wei, Z.","Jin, X.","Tan, X.","Jin, Y.","Chen, E."],"bibdata":{"title":"Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation","type":"inproceedings","year":"2022","pages":"7181-7190","websites":"https://github.com/xiaoachen98/DALN","id":"4cac66c7-fa19-38b6-bae1-d3dcc2048340","created":"2023-01-06T14:07:13.088Z","accessed":"2023-01-06","file_attached":"true","profile_id":"235249c2-3ed4-314a-b309-b1ea0330f5d9","group_id":"5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1","last_modified":"2023-01-13T07:12:08.126Z","read":"true","starred":false,"authored":false,"confirmed":false,"hidden":false,"folder_uuids":"bd3c6f2e-3514-47cf-bc42-12db8b9abe45","private_publication":false,"abstract":"Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .","bibtype":"inproceedings","author":"Chen, Lin and Chen, Huaian and Wei, Zhixiang and Jin, Xin and Tan, Xiao and Jin, Yi and Chen, Enhong","booktitle":"IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022","bibtex":"@inproceedings{\n title = {Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation},\n type = {inproceedings},\n year = {2022},\n pages = {7181-7190},\n websites = {https://github.com/xiaoachen98/DALN},\n id = {4cac66c7-fa19-38b6-bae1-d3dcc2048340},\n created = {2023-01-06T14:07:13.088Z},\n accessed = {2023-01-06},\n file_attached = {true},\n profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},\n group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},\n last_modified = {2023-01-13T07:12:08.126Z},\n read = {true},\n starred = {false},\n authored = {false},\n confirmed = {false},\n hidden = {false},\n folder_uuids = {bd3c6f2e-3514-47cf-bc42-12db8b9abe45},\n private_publication = {false},\n abstract = {Adversarial learning has achieved remarkable performances for unsupervised domain adaptation (UDA). Existing adversarial UDA methods typically adopt an additional discriminator to play the min-max game with a feature ex-tractor. However, most of these methods failed to effectively leverage the predicted discriminative information, and thus cause mode collapse for generator. In this work, we address this problem from a different perspective and design a simple yet effective adversarial paradigm in the form of a discriminator-free adversarial learning network (DALN), wherein the category classifier is reused as a discrimina-tor, which achieves explicit domain alignment and category distinguishment through a unified objective, enabling the DALN to leverage the predicted discriminative information for sufficient feature alignment. Basically, we introduce a Nuclear-norm Wasserstein discrepancy (NWD) that has definite guidance meaning for performing discrimination. Such NWD can be coupled with the classifier to serve as a discrim-inator satisfying the K-Lipschitz constraint without the requirements of additional weight clipping or gradient penalty strategy. Without bells and whistles, DALN compares favorably against the existing state-of-the-art (SOTA) methods on a variety of public datasets. Moreover, as a plug-and-play technique, NWD can be directly used as a generic regular-izer to benefit existing UDA algorithms. Code is available at https://github.com/xiaoachen98/DALN .},\n bibtype = {inproceedings},\n author = {Chen, Lin and Chen, Huaian and Wei, Zhixiang and Jin, Xin and Tan, Xiao and Jin, Yi and Chen, Enhong},\n booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022}\n}","author_short":["Chen, L.","Chen, H.","Wei, Z.","Jin, X.","Tan, X.","Jin, Y.","Chen, E."],"urls":{"Paper":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c/file/4936f385-95f6-5279-18a0-ba2a7e865182/full_text.pdf.pdf","Website":"https://github.com/xiaoachen98/DALN"},"biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","bibbaseid":"chen-chen-wei-jin-tan-jin-chen-reusingthetaskspecificclassifierasadiscriminatordiscriminatorfreeadversarialdomainadaptation-2022","role":"author","metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","dataSources":["2252seNhipfTmjEBQ"],"keywords":[],"search_terms":["reusing","task","specific","classifier","discriminator","discriminator","free","adversarial","domain","adaptation","chen","chen","wei","jin","tan","jin","chen"],"title":"Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation","year":2022}