Add to Chrome
Add to Firefox
Add to EdgeOmni-Scale Feature Learning for Person Re-Identification
May 02, 2019
As an instance-level recognition problem, person re-identification (ReID) relies on discriminative features, which not only capture different spatial scales but also encapsulate an arbitrary combination of multiple scales. We call these features of both homogeneous and heterogeneous scales omni-scale features. In this paper, a novel deep CNN is designed, termed Omni-Scale Network (OSNet), for omni-scale feature learning in ReID. This is achieved by designing a residual block composed of multiple convolutional feature streams, each detecting features at a certain scale. Importantly, a novel unified aggregation gate is introduced to dynamically fuse multi-scale features with input-dependent channel-wise weights. To efficiently learn spatial-channel correlations and avoid overfitting, the building block uses both pointwise and depthwise convolutions. By stacking such blocks layer-by-layer, our OSNet is extremely lightweight and can be trained from scratch on existing ReID benchmarks. Despite its small model size, our OSNet achieves state-of-the-art performance on six person-ReID datasets.
