Dongliang Luo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Dongliang Luo has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Xiang Bai
Huazhong University of Science and Technology, Wuhan, China
6 shared publications - Yuliang Liu
Huazhong University of Science and Technology, Wuhan, China
4 shared publications - Dingkang Liang
Huazhong University of Science and Technology
2 shared publications - Jishen Zeng
Alibaba Group
2 shared publications - Rui Yang
Alibaba Group
2 shared publications - Xianjin Liu
Alibaba Group
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17019-17026
TokBench: Evaluating Your Visual Tokenizer before Visual Generation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17019-17026
Open AccessTokBench: Evaluating Your Visual Tokenizer before Visual Generation
- Open Access
TokBench: Evaluating Your Visual Tokenizer before Visual Generation
In this work, we reveal the limitations of visual tokenizers and VAEs in preserving fine-grained features, and propose a benchmark to evaluate reconstruction performance for two challenging visual contents: text and face. Visual tokenizers and VAEs have significantly advanced visual generation and multimodal modeling by providing more efficient compressed or quantized image representations. However, while helping production models reduce computational burdens, the information loss from image compression fundamentally limits the upper bound of visual generation quality. To evaluate this upper bound, we focus on assessing reconstructed text and facial features since they typically: 1) exist at smaller scales, 2) contain dense and rich textures, 3) are prone to collapse, and 4) are highly sensitive to human vision. We first collect and curate a diverse set of clear text and face images from existing datasets. Unlike approaches using VLM models, we employ established OCR and face recognition models for evaluation, ensuring accuracy while maintaining an exceptionally lightweight assessment process requiring just 2GB memory and 4 minutes to complete. Using our benchmark, we analyze text and face reconstruction quality across various scales for different image tokenizers and VAEs. Our results show modern visual tokenizers still struggle to preserve fine-grained features, especially at smaller scales. We further extend this evaluation framework to video, conducting comprehensive analysis of video tokenizers. Additionally, we demonstrate that traditional metrics fail to accurately reflect reconstruction performance for faces and text, while our proposed metrics serve as an effective complement.
Open AccessTokBench: Evaluating Your Visual Tokenizer before Visual Generation
In this work, we reveal the limitations of visual tokenizers and VAEs in preserving fine-grained features, and propose a benchmark to evaluate reconstruction performance for two challenging visual contents: text and face. Visual tokenizers and VAEs have significantly advanced visual generation and multimodal modeling by providing more efficient compressed or quantized image representations. However, while helping production models reduce computational burdens, the information loss from image compression fundamentally limits the upper bound of visual generation quality. To evaluate this upper bound, we focus on assessing reconstructed text and facial features since they typically: 1) exist at smaller scales, 2) contain dense and rich textures, 3) are prone to collapse, and 4) are highly sensitive to human vision. We first collect and curate a diverse set of clear text and face images from existing datasets. Unlike approaches using VLM models, we employ established OCR and face recognition models for evaluation, ensuring accuracy while maintaining an exceptionally lightweight assessment process requiring just 2GB memory and 4 minutes to complete. Using our benchmark, we analyze text and face reconstruction quality across various scales for different image tokenizers and VAEs. Our results show modern visual tokenizers still struggle to preserve fine-grained features, especially at smaller scales. We further extend this evaluation framework to video, conducting comprehensive analysis of video tokenizers. Additionally, we demonstrate that traditional metrics fail to accurately reflect reconstruction performance for faces and text, while our proposed metrics serve as an effective complement.
- 27 July 2024
- Published by Elsevier in Pattern Recognition
- Vol. 157
Toward real text manipulation detection: New dataset and new solution
- 27 July 2024
- Published by Elsevier in Pattern Recognition
- Vol. 157
Toward real text manipulation detection: New dataset and new solution
- Open Access
A Discrepancy Aware Framework for Robust Anomaly Detection
Defect detection is a critical research area in artificial intelligence. Recently, synthetic data-based self-supervised learning has shown great potential on this task. Although many sophisticated synthesizing strategies exist, little research has been done to investigate the robustness of models when faced with different strategies. In this paper, we focus on this issue and find that existing methods are highly sensitive to them. To alleviate this issue, we present a Discrepancy Aware Framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies across different anomaly detection benchmarks. We hypothesize that the high sensitivity to synthetic data of existing self-supervised methods arises from their heavy reliance on the visual appearance of synthetic data during decoding. In contrast, our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. To this end, inspired by existing knowledge distillation methods, we employ a teacher-student network, which is trained based on synthesized outliers, to compute the discrepancy map as the cue. Extensive experiments on two challenging datasets prove the robustness of our method. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance. Code is available at: https://github.com/caiyuxuan1120/DAF.
Open AccessA Discrepancy Aware Framework for Robust Anomaly Detection
Defect detection is a critical research area in artificial intelligence. Recently, synthetic data-based self-supervised learning has shown great potential on this task. Although many sophisticated synthesizing strategies exist, little research has been done to investigate the robustness of models when faced with different strategies. In this paper, we focus on this issue and find that existing methods are highly sensitive to them. To alleviate this issue, we present a Discrepancy Aware Framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies across different anomaly detection benchmarks. We hypothesize that the high sensitivity to synthetic data of existing self-supervised methods arises from their heavy reliance on the visual appearance of synthetic data during decoding. In contrast, our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. To this end, inspired by existing knowledge distillation methods, we employ a teacher-student network, which is trained based on synthesized outliers, to compute the discrepancy map as the cue. Extensive experiments on two challenging datasets prove the robustness of our method. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance. Code is available at: https://github.com/caiyuxuan1120/DAF.
- 4 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (3) , 3986-3995
A Discrepancy Aware Framework for Robust Anomaly Detection
Defect detection is a critical research area in artificial intelligence. Recently, synthetic data-based self-supervised learning has shown great potential on this task. Although many sophisticated synthesizing strategies exist, little research has been done to investigate the robustness of models when faced with different strategies. In this article, we focus on this issue and find that existing methods are highly sensitive to them. To alleviate this issue, we present a discrepancy aware framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies across different anomaly detection benchmarks. We hypothesize that the high sensitivity to synthetic data of existing self-supervised methods arises from their heavy reliance on the visual appearance of synthetic data during decoding. In contrast, our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. To this end, inspired by existing knowledge distillation methods, we employ a teacher-student network, which is trained based on synthesized outliers, to compute the discrepancy map as the cue. Extensive experiments on two challenging datasets prove the robustness of our method. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance.
- 4 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (3) , 3986-3995
A Discrepancy Aware Framework for Robust Anomaly Detection
Defect detection is a critical research area in artificial intelligence. Recently, synthetic data-based self-supervised learning has shown great potential on this task. Although many sophisticated synthesizing strategies exist, little research has been done to investigate the robustness of models when faced with different strategies. In this article, we focus on this issue and find that existing methods are highly sensitive to them. To alleviate this issue, we present a discrepancy aware framework (DAF), which demonstrates robust performance consistently with simple and cheap strategies across different anomaly detection benchmarks. We hypothesize that the high sensitivity to synthetic data of existing self-supervised methods arises from their heavy reliance on the visual appearance of synthetic data during decoding. In contrast, our method leverages an appearance-agnostic cue to guide the decoder in identifying defects, thereby alleviating its reliance on synthetic appearance. To this end, inspired by existing knowledge distillation methods, we employ a teacher-student network, which is trained based on synthesized outliers, to compute the discrepancy map as the cue. Extensive experiments on two challenging datasets prove the robustness of our method. Under the simple synthesis strategies, it outperforms existing methods by a large margin. Furthermore, it also achieves the state-of-the-art localization performance.
- 19 August 2023
- Published by Springer Nature
- p. 587-600
ICDAR 2023 Competition on Detecting Tampered Text in Images
- 19 August 2023
- Published by Springer Nature
- p. 587-600
ICDAR 2023 Competition on Detecting Tampered Text in Images
- 20 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 777-781
Prior Matching Operator in Self-Supervised Learning
In self-supervised representation learning (SSL), contrastive learning has achieved remarkable successes in recent years. In contrastive learning, the augmented views of the same image are brought closer (i.e., positive pairs), while views from different images (i.e., negative pairs) are separated apart. To perform well, contrastive methods of ten rely on a large number of negative pairs, which is computationally demanding. Intuitively, the negative samples play a role to scatter the representation of samples in space, which guarantees the uniformity of the representation space, thus making it the key to avoiding representation collapse (i.e., constant features). In this work, we propose an interpretable method, in which prior distribution matching is applied to prevent representation collapse, while no negative pair is required. Empirical experiments show promising results of our method yielded by the priors of Gaussian, uniform, and spherical distribution on the CIFAR-10 and CIFAR-100 datasets.
- 20 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 777-781
Prior Matching Operator in Self-Supervised Learning
In self-supervised representation learning (SSL), contrastive learning has achieved remarkable successes in recent years. In contrastive learning, the augmented views of the same image are brought closer (i.e., positive pairs), while views from different images (i.e., negative pairs) are separated apart. To perform well, contrastive methods of ten rely on a large number of negative pairs, which is computationally demanding. Intuitively, the negative samples play a role to scatter the representation of samples in space, which guarantees the uniformity of the representation space, thus making it the key to avoiding representation collapse (i.e., constant features). In this work, we propose an interpretable method, in which prior distribution matching is applied to prevent representation collapse, while no negative pair is required. Empirical experiments show promising results of our method yielded by the priors of Gaussian, uniform, and spherical distribution on the CIFAR-10 and CIFAR-100 datasets.