GraVi-T icon indicating copy to clipboard operation
GraVi-T copied to clipboard

Regarding Action Localization Task

Open deepsworld opened this issue 2 years ago • 1 comments

Hi @kylemin,

Thank you for sharing your excellent work.

Can you please let me know if you use ground truth tracks and detection boxes for AVA dataset?

deepsworld avatar Dec 22 '23 17:12 deepsworld

Thanks for your interest in this work! 36.8% is achieved when using the ground-truth person_ids. If we use a simple tracking algorithm (based only on IoUs) for the person_ids, our method produces 31.0%.

Best, Kyle

kylemin avatar Dec 22 '23 22:12 kylemin