Can masking background and object reduce static bias for zero-shot action recognition?

Re-assign community