Clusformer: A Transformer based Clustering Approach to Unsupervised Large-scale Face and Visual Landmark Recognition

Nguyen, Xuan-Bac; Bui, Duc T.; Duong, Chi Nhan; Bui, Tien D.; Luu, Khoa

doi:10.1109/cvpr46437.2021.01070

Cited by 32 publications

(11 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, there is still a gap in utilizing large unlabelled images and videos to learn Re-ID models in an unsupervised end-to-end manner, which is more practical in real-world applications. Therefore, we suggest there is a great research opportunity in unsupervised endto-end person re-identification, in particular, leveraging the evolutionary vision transformers [123,124,125,126].…”

Section: Discussion and Future Directionsmentioning

confidence: 99%

Unsupervised Person Re-Identification: A Systematic Survey of Challenges and Solutions

Lin¹,

Ren²,

Yeh³

et al. 2021

Preprint

View full text Add to dashboard Cite

Person re-identification (Re-ID) has been a significant research topic in the past decade due to its real-world applications and research significance. While supervised person Re-ID methods achieve superior performance over unsupervised counterparts, they can not scale to large unlabelled datasets and new domains due to the prohibitive labelling cost. Therefore, unsupervised person Re-ID has drawn increasing attention for its potential to address the scalability issue in person Re-ID. Unsupervised person Re-ID is challenging primarily due to lacking identity labels to supervise person feature representation learning. The corresponding solutions are diverse and complex, with various merits and limitations. Therefore, comprehensive surveys on this topic are essential to summarise challenges and solutions to foster future research. Existing person Re-ID surveys have focused on supervised methods from classifications and applications. Still, they lack detailed discussion on how the person Re-ID solutions address the underlying person Re-Id challenges. This survey review recent works on unsupervised person Re-ID from the perspective of challenges and solutions. Specifically, we provide an in-depth analysis of highly influential methods considering the four significant challenges in unsupervised person Re-ID: 1) lacking ground-truth identity labels to supervise person feature learning; 2) learning discriminative person features with pseudo-supervision; 3) learning crosscamera invariant person features and 4) the domain gap between datasets. We summarise and analyze evaluation results and provide insights on the effectiveness of the solutions. Finally, we discuss open issues and suggest some promising future research directions.

show abstract

Section: Discussion and Future Directionsmentioning

confidence: 99%

Unsupervised Person Re-Identification: A Systematic Survey of Challenges and Solutions

Lin¹,

Ren²,

Yeh³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…DA-Net (Guo et al, 2020) conducts clustering by leveraging non-local context information through density-based graph. Clusformer (Nguyen et al, 2021) clusters faces with a transformer. STAR-FC (Shen et al, 2021) develops a structure-preserved sampling strategy to train the edge classification GCN.…”

Section: Related Workmentioning

confidence: 99%

“…To satisfy the two principles, the criterion is proposed according to the F β -score (Rijsbergen, 1979) in information retrieval. Similar to visual grammars (Nguyen et al, 2021), all candidate neighbours are ordered by the similarity with the probe vertex in a sequence. Given candidate neighbours of size j probed by vertex v i , its quality criterion Q (j) is defined as:…”

Section: Candidate Neighbours Quality Criterionmentioning

confidence: 99%

“…The experiments are conducted with PyTorch (Paszke et al, 2019) and DGL (Wang et al, 2019a). (Lloyd, 1982), HAC (Sibson, 1973), DBSCAN (Ester et al, 1996), and graph-based methods L-GCN (Wang et al, 2019b), DS-GCN , VE-GCN , DA-Net (Guo et al, 2020), Clusformer (Nguyen et al, 2021) and STAR-FC (Shen et al, 2021). In this section, to further enhance the clustering performance of GCNs, some noise is added to the training graph.…”

Section: Evaluation Metrics Datasets and Experimental Settingsmentioning

confidence: 99%

“…Too many edges connected will increase the number of noise edges, and the vertex feature will be polluted by wrongly connected vertices. Although Clusformer (Nguyen et al, 2021) and GAT (Velickovic et al, 2018) try to reduce the impact of the noise edges by the attention mechanism, the connections between various vertices are very complex, and thus it is difficult to find common patterns for the attention weight learning .…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Ada-NETS: Face Clustering via Adaptive Neighbour Discovery in the Structure Space

Wang¹,

Zhang²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Face clustering has attracted rising research interest recently to take advantage of massive amounts of face images on the web. State-of-the-art performance has been achieved by Graph Convolutional Networks (GCN) due to their powerful representation capacity. However, existing GCN-based methods build face graphs mainly according to kNN relations in the feature space, which may lead to a lot of noise edges connecting two faces of different classes. The face features will be polluted when messages pass along these noise edges, thus degrading the performance of GCNs. In this paper, a novel algorithm named Ada-NETS is proposed to cluster faces by constructing clean graphs for GCNs. In Ada-NETS, each face is transformed to a new structure space, obtaining robust features by considering face features of the neighbour images. Then, an adaptive neighbour discovery strategy is proposed to determine a proper number of edges connecting to each face image. It significantly reduces the noise edges while maintaining the good ones to build a graph with clean yet rich edges for GCNs to cluster faces. Experiments on multiple public clustering datasets show that Ada-NETS significantly outperforms current state-of-the-art methods, proving its superiority and generalization.

show abstract

On Mitigating Hard Clusters for Face Clustering

Chen

Zhong

Chen

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Clusformer: A Transformer based Clustering Approach to Unsupervised Large-scale Face and Visual Landmark Recognition

Cited by 32 publications

References 25 publications

Unsupervised Person Re-Identification: A Systematic Survey of Challenges and Solutions

Unsupervised Person Re-Identification: A Systematic Survey of Challenges and Solutions

Ada-NETS: Face Clustering via Adaptive Neighbour Discovery in the Structure Space

On Mitigating Hard Clusters for Face Clustering

Contact Info

Product

Resources

About