Full text
A Deep Learning Loss Based on Additive Cosine Margin: Application to Fashion Style and Face Recognition Pendar Alirezazadeh University of the Basque Country UPV/EHU, San Sebastian, Spain Fadi Dornaika University of the Basque Country UPV/EHU, San Sebastian, Spain Ho Chi Minh City Open University, Vietnam Abdelmalik Moujahid University of the Basque Country UPV/EHU, San Sebastian, Spain Published in: Applied Soft Computing 131 (2022) 109776, Elsevier. DOI: 10.1016/j.asoc.2022.109776. Accepted October 29, 2022. Abstract Recently, loss functions based on angular spans improved the performance of deep visual recognition. These losses converted Euclidean cross entropy to angular cross entropy loss. Fashion style recognition deals with the problem of assigning a person’s outfit to a fashion style category. Due to the high similarity between different clothing items and the use of softmax-based loss functions, many current methods that address this problem show relatively poor performance and cannot guarantee sufficient inter-class margins in the fashion domain. In this work, we propose an end-toend method for deep visual recognition by combining a standard CNN architecture with a novel loss function, which we call Additive Cosine Margin Loss (ACML). The proposed function not only projects feature vectors of different classes into different regions of the embedding, but also enforces compactness of the projections within each class. Our experiments were conducted on two public and well-known fashion style recognition datasets (FashionStyle14 and HipsterWars), and on the face verification and identification datasets (LFW, YTF, and MegaFace). These experiments demonstrate the superiority of the proposed loss function over (i) existing angular margin-based loss functions, and (ii) state-of-the-art methods for clothing style recognition as well as face analysis tasks. Main Contributions •Proposes a novel loss function called Additive Cosine Margin Loss (ACML), which jointly enforces inter-class separation and intra-class compactness by adding an angular mini-margin to the cosine margin in deep learning architectures. •Demonstrates that the ACML loss is more effective than traditional margin-based losses (SphereFace, AM-Softmax, ArcFace) on style and face recognition datasets, improving global and per-class accuracy in empirical benchmarking. 1
•Shows through ablation and benchmarking that ACML yields more balanced performance across classes and tasks, outperforming state-of-the-art deep learning methods in fashion and face recognition, including on challenging benchmarks like MegaFace. •Presents theoretical and geometric analysis illustrating how ACML achieves both intra-class compactness and inter-class discrimination, giving more robust feature embeddings and decision margins than competitors. •Empirically validates ACML on FashionStyle14, HipsterWars, LFW, YTF, and MegaFace datasets, achieving up to 90% accuracy on fashion recognition and state-of-the-art results in face verification and identification tasks. Reference: Pendar Alirezazadeh, Fadi Dornaika, Abdelmalik Moujahid. “A deep learning loss based on additive cosine margin: Application to fashion style and face recognition.” Applied Soft Computing, Volume 131, 2022, 109776. DOI: 10.1016/j.asoc.2022.109776 2
