List-wise learning-to-rank with convolutional neural networks for person re-identification

Yiqiang Chen (1) , Stefan Duffner (1) , Andrei Stoian (2) , Jean-Yves Dufour (2) , Atilla Baskurt (1)

1 imagine - Extraction de Caractéristiques et Identification
2 Thales Services

In this paper, we present a novel machine learning-based image ranking approach using Convolutional Neural Networks (CNN). Our proposed method relies on a similarity metric learning algorithm operating on lists of image examples and a loss function taking into account the ranking in these lists with respect to different query images. This comprises two major contributions: (1) Rank lists instead of image pairs or triplets are used for training, thus integrating more explicitly the order of similarity and relations between sets of images. (2) A weighting is introduced in the loss function based on two evaluation measures: the mean average precision and the rank 1 score. We evaluated our approach on two different computer vision applications that are commonly formulated as ranking problems: person re-identification and image retrieval with several public benchmarks and showed that our new loss function outperforms other common functions and that our method achieves state-of-the-art performance compared to existing approaches from the literature.

Mots clés: similarity metric learning; learning-to-rank; person re-identification; Deep Learning; image retrieval

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]; Réseau de neurones [cs.NE]; Traitement des images [eess.IV]

Yiqiang Chen, Stefan Duffner, Andrei Stoian, Jean-Yves Dufour, Atilla Baskurt. List-wise learning-to-rank with convolutional neural networks for person re-identification. Machine Vision and Applications, 2021, 32 (2), ⟨10.1007/s00138-021-01170-0⟩. ⟨hal-03157567⟩