Hi , in the paper in methodology you have mentioned / cited a TSN feature which is pre-trained on Kinetics, however, in implementation details , you claim that for fair comparsion you have used same TSN feature as that of BMN/BSN . But BMN/BSN feature is fine-tuned on ActivityNet. WHich one is correct ?
Hi, thanks for your attention to SSTAP. I3D features are pre-trained on Kinetics and without fine-tuned on ActivityNet v1.3. TSN features are consistent with BMN/BSN.
Hi , in the paper in methodology you have mentioned / cited a TSN feature which is pre-trained on Kinetics, however, in implementation details , you claim that for fair comparsion you have used same TSN feature as that of BMN/BSN . But BMN/BSN feature is fine-tuned on ActivityNet. WHich one is correct ?