Robust Re-Identification by Multiple Views Knowledge Distillation

Porrello, Angelo; Bergamini, Luca; Calderara, Simone

doi:10.1007/978-3-030-58607-2_6

Cited by 50 publications

(34 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Contrary to I2I and V2V Re-ID, I2V Re-ID targets to build bridges between image representations and video representations. Recent works [1,2] formulate this problem as a knowledge distillation process. Specifically, Gu et al [1] transfer temporal information from a video-based teacher network to an imagebased student one; Porrello et al [2] distill the multiple views knowledge from teacher to student network.…”

Section: Introductionmentioning

confidence: 99%

“…Recent works [1,2] formulate this problem as a knowledge distillation process. Specifically, Gu et al [1] transfer temporal information from a video-based teacher network to an imagebased student one; Porrello et al [2] distill the multiple views knowledge from teacher to student network. In terms of loss function for distillation, in TKP [1], feature distance and cross sample distance are used for propagation; in VKD [2], logits and pairwise distance are adopted for distillation.…”

Section: Introductionmentioning

confidence: 99%

“…Specifically, Gu et al [1] transfer temporal information from a video-based teacher network to an imagebased student one; Porrello et al [2] distill the multiple views knowledge from teacher to student network. In terms of loss function for distillation, in TKP [1], feature distance and cross sample distance are used for propagation; in VKD [2], logits and pairwise distance are adopted for distillation. All of these distillation losses focus on the global matching, designed for classification tasks [3] or representative learning [4].…”

Section: Introductionmentioning

confidence: 99%

“…Instead of directly comparing the distance between anchor-positive and anchor-negative pairs, we propose to measure the probabilities of these distances. Unlike other losses [1,2] that match the global information between teacher and student, the TCL loss focuses on the local structure consisting of nearest neighbors for each example. Coupled with hard triplet sampling in the student embedding, the discriminative ability of features are enhanced by the soft supervision of the teacher network.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Image-to-Video Re-Identification via Mutual Discriminative Knowledge Transfer

Wang¹,

Wang²,

Li³

2022

Preprint

View full text Add to dashboard Cite

The gap in representations between image and video makes Image-to-Video Re-identification (I2V Re-ID) challenging, and recent works formulate this problem as a knowledge distillation (KD) process. In this paper, we propose a mutual discriminative knowledge distillation framework to transfer a video-based richer representation to an image based representation more effectively. Specifically, we propose the triplet contrast loss (TCL), a novel loss designed for KD. During the KD process, the TCL loss transfers the local structure, exploits the higher order information, and mitigates the misalignment of the heterogeneous output of teacher and student networks. Compared with other losses for KD, the proposed TCL loss selectively transfers the local discriminative features from teacher to student, making it effective in the ReID. Besides the TCL loss, we adopt mutual learning to regularize both the teacher and student networks training. Extensive experiments demonstrate the effectiveness of our method on the MARS, DukeMTMC-VideoReID and VeRi-776 benchmarks.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Image-to-Video Re-Identification via Mutual Discriminative Knowledge Transfer

Wang¹,

Wang²,

Li³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Aiming to search the identity captured by a set of images, image set retrieval has been a crucial problem in recent years, especially due to the rapid development of camera surveillance system, and the availability of multiple images of the same identity. There exists many practical applications for image set retrieval such as video face recognition [1,2], multi-shot person re-identification [3,4,5,6] and gait recognition [7,8]. Comparing to a single image, a set of images usually contains This research was supported in part by DARPA grant FA8750-18-2-0018 funded under the AIDA program and the DARPA KAIROS program from the Air Force Research Laboratory under agreement number FA8750-19-2-0200.…”

Section: Introductionmentioning

confidence: 99%

Statistical Distance Metric Learning for Image Set Retrieval

Hauptmann

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Measuring similarity between two image sets is instrumental in many computer vision tasks, such as video face recognition, multi-shot person re-identification and gait recognition. In most of the recent works, it is done by aggregating the embedding features of images as a fixed size vector, and calculating a metric in vector space (i.e. Euclidean distance). The embedding feature function can be learned by deep metric learning (DML) technique. However, methods relying on feature aggregation fail to capture the diversity and uncertainty within image sets. In this paper, we obviate the need of feature aggregation and propose a novel Statistical Distance Metric Learning (SDML) framework, which represents each image set as a probability distribution in embedding feature space and compares two image sets by statistical distance between their distributions. Among all types of statistical distance, we choose Jeffrey's divergence (JD), which can be obtained from two embedding feature sets by kNN based density estimator. We also design a statistical centroid loss function to enhance the discriminative power of training process. Our SDML framework naturally preserves the diversity within an image set, and the relation between two sets. We evaluate our proposed approach on gait recognition and multi-shot person re-id. The experiment results show that SDML outperforms conventional DML, and also receives competitive/superior performance comparing to the previous state-of-the-arts on the aforementioned tasks.

show abstract