当前位置:网站首页>The generalization gap (CS CV) of one-time target detection is eliminated

The generalization gap (CS CV) of one-time target detection is eliminated

2020-12-07 19:23:48 Ling Qian

Although substantial progress has been made in target detection and less shot learning , But detection objects based on a single instance —— Single shot target detection —— It's still a challenge : The training model shows a huge promotion gap , Among them, the object class used during training is more reliable than the new class detection . Here we show that , This generalization gap can be closed by increasing the number of object categories used in training . It turns out that , The model realizes the transformation from memorizing a single category to the similarity of learning objectives in the category distribution , It has strong generalization ability in testing . It is important to , In this system , A standard method to improve the target detection model , Like a stronger backbone or a longer training program , It's also good for new categories , This is not COCO In the case of such a small data set . Our results show that , The key of strong and few shot detection model may not be the complex metric learning method , It's about the number of extended categories . therefore , Future data annotation work should focus on a broader dataset , And more categories , Instead of collecting more images or instances per category .

Original title :Closing the Generalization Gap in One-Shot Object Detection

original text :Despite substantial progress in object detection and few-shot learning, detecting objects based on a single example - one-shot object detection - remains a challenge: trained models exhibit a substantial generalization gap, where object categories used during training are detected much more reliably than novel ones. Here we show that this generalization gap can be nearly closed by increasing the number of object categories used during training. Our results show that the models switch from memorizing individual categories to learning object similarity over the category distribution, enabling strong generalization at test time. Importantly, in this regime standard methods to improve object detection models like stronger backbones or longer training schedules also benefit novel categories, which was not the case for smaller datasets like COCO. Our results suggest that the key to strong few-shot detection models may not lie in sophisticated metric learning approaches, but instead in scaling the number of categories. Future data annotation efforts should therefore focus on wider datasets and annotate a larger number of categories rather than gathering more images or instances per category.

Original author :Claudio Michaelis, Matthias Bethge, Alexander S. Ecker Original address :https://arxiv.org/abs/2011.04267

Original statement , This article is authorized by the author + Community publication , Unauthorized , Shall not be reproduced .

If there is any infringement , Please contact the yunjia_community@tencent.com Delete .

版权声明
本文为[Ling Qian]所创,转载请带上原文链接,感谢
https://chowdera.com/2020/11/20201119031742514i.html