There is no need to consider the data set. You should find a data set about the problem you want to solve on the Internet. If you want to make your own data set, it will be very troublesome. Just read a few papers.

There are many online tutorials about training your own data sets with v5. The code of v5 is very convenient and modular.

It is difficult to understand the principle of v5 and how to add some latest algorithms in v5 to improve the effect of your training set. If you are good enough, you can also propose your own new algorithm for a section of v5, such as improving CIOU.

Literature review should be well written and the paper format should be well done.