VSEImageEncoder - host ourself

[florian-hoenicke]

Building the VSEImageEncoder causes some issues from time to time as pointed out by @nan-wang

• it is running out of disk. We can add RUN du -fh to check the disk usage for each step
• The model downloading is not working. https://github.com/jina-ai/jina-hub/runs/1975233650?check_suite_focus=true#step:6:414
• The logs from wget blew up the logging outputs. We need wget -q to make it viewable.

https://github.com/jina-ai/jina-hub/issues/4420

The reason is that we have to download an archive which is quite big. However, we only need one of these files. Also, the upstream where the file is hosted is limited.

Luckily, the owner allowed us to mirror the model. Thanks @fartashf!

UAC:

1. implementation
   • download large file
   • upload relevant model to a mirror
   • make the VSEImageEncoder use the new mirror
2. add references to
   • the original models
   • the paper
   • github repo
3. add the links of the model mirror to the readme of this repo
   • https://github.com/fartashf/vsepp/issues/38

[JoanFM]

@FionnD @lusloher @florian-hoenicke what should we do with this issue?

[nan-wang]

Let's host the model using git-lfs