Pretrained issue

[Joao-L-S-Almeida]

It automatically sets pretrained=False when no config is available (as for Swin).

[CarlosGomes98]

I think its safer to throw an error here and force the user to specify pretrained=False. What we can do is have the error message ask the user to do just that. In this way, the user is crystal clear on what weights area actually being loaded (or not) in the model.

[romeokienzler]

@Michal-Muszynski what is the reason you can't update the config file?

[biancazadrozny]

Need to test if during inference this error occurs when using a fine-tuned model and the backbone is not there, with this flag.

[Joao-L-S-Almeida]

WARNING. If this PR isn't approved (or rejected) within 7 days it will be approved as is.

[romeokienzler]

@romeokienzler please review

[romeokienzler]

@Michal-Muszynski does this solve his problem?

[romeokienzler]

@Joao-L-S-Almeida I think there is one little problem, there rest lgtm thanks a lot!