How to infer with my own trained model?

[hyclover]

How to infer with my own trained model?

[DrSleep]

You first need to define the model you trained (i.e. MBv2, or RF-LW-50/101/152) with the appropriate number of classes.
In the notebook I show how to do it for each network, using this: net = fun(n_classes, pretrained=True).eval().cuda().

You should replace fun with the appropriate network function and set pretrained=False
Then you need to load the checkpoint file, e.g. via ckpt = torch.load(checkpoint)
If you trained your model with GPU, then you will need to do net = nn.DataParallel(net) before restoring the state dict net.load_state_dict(ckpt["model"]) (if you were not training with GPU, you can just skip the first step)

You can also see how I am loading the checkpoint here:

light-weight-refinenet/src_v2/train.py

Line 40 in b6871c2

def setup_checkpoint_and_maybe_restore(args, model, optimisers, schedulers):

[hyclover]

Thanks for your reply! I have achieved it successfully according to your guidance.

[DrSleep]

Marking it as resolved