Want to create pull request to make multi GPU training more robust

[fhandke-fugro]

Hello,

when running the training on a multi GPU system within a docker container I had some issue with processes trying to access files that were still being written to by the main process (namely when downloading the pre-trained weights and when accessing the stage 1 'best_stg1.pth' weights before continuing with stage 2.

I made changes to the files /src/nn/backbone/hgnetv2.py and /src/solver/det_solver.py and would like to open a pull request for those changes. In general they should robustify the training and avoid running conditions.

Could you let me know how I could open a pull request? Because I do not seem to be able to push a new branch to the repo.

Thanks