question
Closed this issue · 8 comments
I want to train a new model, do I need to get his masking map first?
This is an unsupervised model, no labeling needed for training.
I don't need this picture？
Only for evaluation, training is done merely on the input images.
Then I want to extract the video background of multiple people, can I just use tennis.ckpt?
no, the model should be trained per scene, if your video takes place in a different scene you should train a new model and use the new set of weights.
I highly recommend reading the paper to better understand the idea of this background model.
ok，thank
When I used my new weights, I reported this error "raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for BMN:", this is why
Can you please give more details to reproduce the error?