OpenGVLab / unmasked_teacher

[ICCV2023 Oral] Unmasked Teacher: Towards Training-Efficient Video Foundation Models
https://arxiv.org/abs/2303.16058
MIT License
285 stars 15 forks source link

Clarity on retrieval results #1

Closed ffnc1020 closed 1 year ago

ffnc1020 commented 1 year ago

Hi,

Thanks for the impressive results! I want to have some clarity on the reported retrieval results:

  1. Since you are using VTM loss, during retrieval, do you re-rank the candidates with VTM score among top candidates?
  2. Are the reported number using duel softmax loss and QB-Norm? Thanks!
Andy1621 commented 1 year ago

Sorry for the late reply.

  1. Yes, we follow ALBEF and Singularity to re-rank the candidates.
  2. We do not use other post-processing techniques like QB-Norm.
ffnc1020 commented 1 year ago

Thanks for the clarification!