Model training method, pedestrian re-identification method and device

Abstract

The invention discloses a model training method, a pedestrian re-identification method and a pedestrian re-identification device, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: inputting a training sample set into a feature extraction network for feature extraction to obtain a first feature map; inputting the first feature maps belonging to the same training sample into a feature pyramid network for up-sampling, and in the up-sampling process, fusing second feature maps output by a plurality of second feature extraction layers of the feature pyramid network with the first feature map of the previous level through a variable convolutional layer to obtain a fused feature map; and performing pedestrian target detection and identity recognition on the fused feature map to obtain a corresponding pedestrian position and a pedestrian identity, and performing parameter adjustment on the feature extraction network and the feature pyramid network through a preset loss function to complete model training. According to the embodiment, anchor-frame-free and single-step-strategy target detection is realized, the calculation amount is reduced, and meanwhile, multi-scale features can be better aggregated.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a model training method, a pedestrian re-identification method and a device. Background technique [0002] Pedestrian re-identification aims to identify the location and identity of pedestrians in videos or images at the same time. There are usually two implementation methods in the existing technology. One is a two-step strategy, that is, first detect the target, locate the pedestrian's position, extract the pedestrian from the video or image, and then perform feature extraction through the subsequent network to identify the pedestrian. Identity, a typical network is Faster-RCNN (an object detection algorithm). The other is a one-step strategy, which realizes object detection and identification of pedestrians end-to-end. Specifically, the anchor frame is used to detect the target first, and the features extracted during the target detection process are shared with the subseq...

AI Technical Summary

Problems solved by technology

In addition, in the prior art, there is also a single-step scheme that uses a pyramid structure instead of an anchor frame for target detection, such as the FCOS (Fully Convolutional One-Stage Object Detection) network. This scheme can detect targets of different scales for the same category, and subsequent use When the query sample is retrieved in the database, there is a problem of scale misalignment, that is, only the results of the same scale as the query sample can be queried, and the results of other scales cannot be queried, resulting in low query accuracy

Images