I'm using this emernerf architecture for my maneuver detection project to get the decomposed field as my input features, thus I am using it as a backbone. I trained it for the 1000 epochs on nuscenes on scene 0 but when I ran it in the eval model on the different scene, the state dict of the saved checkpoint and new model which is supposed to adjust to the new scene had different sizes.
Do we need to train the model for each scene separately? How about it's inference on some custom data using just 6 cameras?
I'm using this emernerf architecture for my maneuver detection project to get the decomposed field as my input features, thus I am using it as a backbone. I trained it for the 1000 epochs on nuscenes on scene 0 but when I ran it in the eval model on the different scene, the state dict of the saved checkpoint and new model which is supposed to adjust to the new scene had different sizes.
Do we need to train the model for each scene separately? How about it's inference on some custom data using just 6 cameras?