Extraction of Camera Movements

[mumucc1]

Great works！ I noticed in your paper that the training set uses a self-collected dataset and extracts the camera parameters from it. In the testing set, however, the realestate10K dataset is used. Could you please clarify whether the method for extracting camera parameters is the same for both datasets? If they are different, how do you ensure the effectiveness of the testing? Thanks！

[WanquanF]

We did not train on the realestate10K dataset, we only used it for testing (generalization only).

I think the method to deal with data should be the same for all datasets.