A Robust Training Method for Object Detection Networks

Abstract

The invention relates to a robust training method for a target detection network, comprising: acquiring training samples, on which some detection targets carry artificial annotation frames; using a target detection network to perform feature extraction on the training samples, and in the training samples A suggestion frame is generated on the training sample; the original sampling label is marked on the suggestion frame, and the original sampling label includes a positive label and a negative label; a pooling branch is used to perform a pooling operation on the positive label, and the first region of interest feature is output ; Input the feature of the first region of interest into a mining network, the mining network is a fully connected neural network, and the mining network generates a new suggestion box label—the mining label; The mining label and the original sampling label are fused to generate gold label; use the gold label for the training of the object detection network.

Description

technical field [0001] The invention relates to the technical field of computer vision and target detection, in particular to a robust training method for target detection network. Background technique [0002] In recent years, object detection frameworks based on Convolutional Neural Networks (CNNs) have become powerful methods for various computer vision tasks, and have been widely used in object localization and object statistics tasks. At the same time, object detection frameworks based on Convolutional Neural Networks (CNNs) have been continuously improved and many excellent architectures have been proposed. Among them, region-based detection frameworks (e.g., FasterRCNN, FPN), which include a preprocessing step for region proposals, are widely used due to their more accurate detection performance. Meanwhile, many methods continue to improve their performance by optimizing the network architecture of feature extractors. However, how to enhance the training robustness ...

AI Technical Summary

Problems solved by technology

[0005] Compared with the existing technology, a robust training method for the target detection network in the embodiment of the present application is adopted, and the process of mining the suggestion frame and label fusion is added in the training process of the target detection network, which effectively overcomes the problem caused by manual labeling. If the frame is missing or the set threshold (the first threshold and the second threshold) is too high or too low, the suggested frame is labeled incorrectly or the phenomenon of too many false positives occurs in the sample, which improves the anti-interference ability of the network training process

Images