Add from_checkpoint method that directly infers the model class

[omkar-334]

Tries to solve #651.
Now, we can load models using

import rfdetr

model_path = "rfdetr_model.pth"
model = rfdetr.from_checkpoint(model_path)

[ashwin3005]

hi @omkar-334 ,

are we trying to infer the model class from the "file name" of the saved weights?

[omkar-334]

hi @ashwin3005 , we are trying to extract the model size from pretrained_weights argument

[ashwin3005]

yeah, but you are comparing predefined names against weights_name. which is a string.

weights_name = str(args.pretrain_weights).lower()

what if the model weights are saved using a different name?