Mmdetection valueerror: need at least one array to concatenate solution

stay mmdetection Sometimes the training model will appear ValueError: need at least one array to concatenate Error of , The details are shown in the figure below .

A lot of people say it's mmdet/dataset Next coco.py or voc.py in CLASSES The setting does not correspond to the dataset , Bloggers don't think so . Because in mmdetection2 in , You don't need to be there mmdet/dataset Revision in China CLASSES, Just set it in the training configuration file .config The interpreter will automatically modify through inheritance CLASSES Value , See the figure below for details , Or reference mmdetection2 The use of tutorials from data processing 、 Profiles to training and testing （ Support coco Data and pascal_voc data ）_ A flash of hope to my blog -CSDN Blog _mmdetection2https://hpg123.blog.csdn.net/article/details/124617894

  As long as the configuration file classes The setting is correct ,voc The above error reports will not appear in the data .

however ,coco The data set is set incorrectly during data conversion , The above error occurs . Bloggers' datasets are only shoot A category , During dataset conversion , In the following ways tools\dataset_converters\pascal_voc.py File set category .【 Wrong way 】

  And then generated coco Data problems , There was only one shoot Categories of coco data , It turns out  20 More than a supercategory project . This is the root cause of error reporting .

When converting datasets , Need to pass in tools\dataset_converters\pascal_voc.py The file modifies the custom category according to the following standards and settings

  Generated in this way coco The data set is strictly correct , Its categories The format of is shown in the figure below . If it is generated through a third-party project coco data , Please check categories Whether the data in and the category of the dataset are normal .

  In the revision coco Data set error , Now? mmdetection You can train data normally .