We show model results on TVQA val and test-public set below. The results are shown in two different tables, depend on whether they use ground-truth timestamp annotation at inference. To participate, submit your results to Codalab w/o ts Evaluation Portal for model without using ground-truth timestamps at inference, or Codalab w/ ts Evaluation Portal for model using ground-truth timestamps at inference. To evaluate TVQA+, follow the instructions here: TVQA+ evaluation portal. Please see most up-to-date leaderboard results here TVQA w/o ts, TVQA w/ ts. TVQA+.
Send me an email (jielei [at] cs.unc.edu) if you want to show your results here.