What is the problem this feature will solve?
I am trying to run inference on new videos using BMN or BSN, but the documentation wants me to first train TSN feature extractor using activitynet myself.
Is it possible to add an already fine-tuned TSN on activitynet to the model zoo? Or can I run inference directly through some already made pipeline by another user? Can I use mmaction.api to directly infer without a feature map?
What is the feature?
A direct inference pipeline that works on new videos for BMN or BSN.
What alternatives have you considered?
No response