A little difference between classification and detection tasks (my humble opinion, please correct me if I am wrong)
The imagenet dataset is mainly used for classification tasks. For multiple objects in a certain picture, its hair type only belongs to a certain object, and it is only a classification for a certain object. It only classifies the main objects in the picture, because the classification cannot be used for detection, and the pure classification network can only do single classification. If you want to classify more, you need to add detection network branches. The data set used for detection tasks in imagenet is relatively small.
The coco data set is mainly used for detection and cannot be used to train a simple classification network (because there is no label for a simple classification network), and can only be used for multi-classification. Multi-classification actually implies the need for a detection network.
————————————————
Copyright statement: This article is an original article of CSDN blogger "qinglv1", following the CC 4.0 BY-SA copyright agreement, please attach the original source link and this statement for reprinting .
Original link: https://blog.csdn.net/qinglv1/article/details/86605378