VisNet: Efficient Person Re-Identification via Alpha-Divergence Loss, Feature Fusion and Dynamic Multi-Task Learning

TL;DR

VisNet tackles efficient person re-identification by fusing multiscale CNN features with learned per-scale attention, guided by semantic clustering using rule-based pseudo-labels and a dynamic, multi-task loss that includes the FIDI metric learning term. The approach achieves competitive Market-1501 results (Rank-1 $87.05\%$, mAP $77.65\%$) with a lightweight footprint (32.41M parameters, 4.601 GFLOPs), offering strong potential for real-time deployment on constrained devices. Key contributions include a scalable, scale-aware fusion mechanism, semantic regularization without teacher-student distillation, and a semantic-aware augmentation pipeline that reduces background reliance. Overall, VisNet demonstrates that carefully designed CNN-based multiscale processing and lightweight semantic supervision can approach Transformer-based accuracy while preserving practical efficiency for surveillance and mobile applications.

Abstract

Person re-identification (ReID) is an extremely important area in both surveillance and mobile applications, requiring strong accuracy with minimal computational cost. State-of-the-art methods give good accuracy but with high computational budgets. To remedy this, this paper proposes VisNet, a computationally efficient and effective re-identification model suitable for real-world scenarios. It is the culmination of conceptual contributions, including feature fusion at multiple scales with automatic attention on each, semantic clustering with anatomical body partitioning, a dynamic weight averaging technique to balance classification semantic regularization, and the use of loss function FIDI for improved metric learning tasks. The multiple scales fuse ResNet50's stages 1 through 4 without the use of parallel paths, with semantic clustering introducing spatial constraints through the use of rule-based pseudo-labeling. VisNet achieves 87.05% Rank-1 and 77.65% mAP on the Market-1501 dataset, having 32.41M parameters and 4.601 GFLOPs, hence, proposing a practical approach for real-time deployment in surveillance and mobile applications where computational resources are limited.

Figures (4)

• Figure 1: Market-1501 Query: The proposed model successfully re-identifies the person (ID:0921) across multiple viewpoints in the top-5 ranked results among 19,733 images.
• Figure 2: Training pipeline of the proposed method.
• Figure 3:
• Figure 4: Spatial Clustering