Learnable Adaptive Cosine Estimator (LACE) for Image Classification

Peeples, Joshua; McCurley, Connor H.; Walker, Sarah; Stewart, Dylan; Zare, Alina

doi:10.1109/wacv51458.2022.00381

Cited by 2 publications

(3 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To fully consider the noises and outliers of skeleton data, motivated by related fine-grained tasks [26], [27], we propose a Robust Decouple Loss (RDL). RDL achieves large class-wise discriminative embedding by leveraging the ratio of norms of each class along with both intra-class and inter-class cosine of the vectors, which significantly improves the robustness of the existing explicit loss function.…”

Section: Arxiv:230615321v1 [Cscv] 27 Jun 2023mentioning

confidence: 99%

“…However, the empirically scalable and marginal hyperparameters limit the discriminant capability of implicit loss. Recent methods [27], [50], [51] employ adaptive hyperparameters to improve the robustness. However, all these implicit methods have not noticed the classwise discriminant on both angular and norm-based aspects, especially for large intra-class and slight inter-class variance of samples.…”

Section: B Loss Functions For Fine-grained Tasksmentioning

confidence: 99%

“…(2) RDL achieves an accuracy of 93.18%, which surpasses the state-of-the-art implicit losses in recent years, including competitive arcFace loss (+0.47%). To further illustrate the Methods FSD-10(%) FineGym99(%) Baseline 91.06 90.13 center loss [19] 91.06 91.17 sphereFace [22] 88.24 -LMCL [49] 92.24 -arcFace [24] 92.71 91.52 LACE [27] 91 effectiveness of RDL, we involve the competitive arcFace and recent LACE for comparison on the FineGym99 dataset. RDL outperforms arcFace and LACE by 0.59% and 0.37%, respectively.…”

Section: Comparison With Other Loss Functionsmentioning

confidence: 99%

See 2 more Smart Citations

Spatial-Temporal Attention Network with Multi-similarity Loss for Fine-Grained Skeleton-Based Action Recognition

Liu

et al. 2021

Neural Information Processing

View full text Add to dashboard Cite

Graph convolutional networks have been widely used in skeleton-based action recognition. However, existing approaches are limited in fine-grained action recognition due to the similarity of inter-class data. Moreover, the noisy data from pose extraction increases the challenge of fine-grained recognition. In this work, we propose a flexible attention block called Channel-Variable Spatial-Temporal Attention (CVSTA) to enhance the discriminative power of spatial-temporal joints and obtain a more compact intra-class feature distribution. Based on CVSTA, we construct a Multi-Dimensional Refinement Graph Convolutional Network (MDR-GCN), which can improve the discrimination among channel-, joint-and frame-level features for fine-grained actions. Furthermore, we propose a Robust Decouple Loss (RDL), which significantly boosts the effect of the CVSTA and reduces the impact of noise. The proposed method combining MDR-GCN with RDL outperforms the known stateof-the-art skeleton-based approaches on fine-grained datasets, FineGym99 and FSD-10, and also on the coarse dataset NTU-RGB+D X-view version.

show abstract

Section: Arxiv:230615321v1 [Cscv] 27 Jun 2023mentioning

confidence: 99%