1 . How to deal with those pyx and.C.H files
Some files under lib are.Pyx files, encounter the file that can not be imported cython, and then make the Lib folder again.
The operation of.C and.H is the same.
2 . It’s best not to use pretrained_model when training your own data, because the type of training varies, loss = inf and loss = nan, – Nan may occur.
3 . Data source checking:
When you’re making your own VOC format data source, check the contents of the. XML file to see if it’s consistent with groudtruth, or the training won’t converge. (prevent undetected errors when generating.Xml from labels.txt)
As for the production of labels, there are many open source make_rec codes on the Internet.
4 . You can not directly use.Sh training, use pycharm or other IDE with debug function to debug first. We have a basic understanding of the flow of data and find the root of the problem at first.
Other common online searches can not be repeated.