* Clear VRAM of face analyser on post process

* Mark as NEXT

* Reduce tensorflow memory to 512 MB

* Cosmetics on installer

* Add is_download_done to pre_process() hook to prevent errors

* Use latest onnxruntime

* Testing for download methods, Make get_download_size more robust

* Testing for download methods

* Introduce --skip-download argument

* Catch exception causes by a firewall

* Looks stable to me