Closed YoungJT-7274 closed 1 year ago
The Gaussian weighted fusion is employed to produce a high-quality action instance which is a tuple (start time, end time, action score). During Training, this action tuple can not be directly used and needs to be converted back to snippet-wise pseudo-label by Linpro pseudo-label generation. During testing, it is adopted after action segment proposals are raised and generates the final result.
Thanks for your reply.
I can't tell the difference between Gaussian weighted fusion and Gaussian weighted fusion (testing) in the paper. Can you tell me where the Gaussian weighted fusion strategy is adopted as post-processing during testing in your code? Thanks a lot.