Track 1: is post-keyframe data allowed, and how are winning entries verified?
#5
by kaelan-yim-pto - opened
Hi organizers, two clarifying questions about Track 1:
- The PAI-AV clips include data after the event keyframe (later video frames and the full ego trajectory). Is any post-keyframe information allowed as model input in Track 1, or only data up to the keyframe?
- Since the board can now reach near-ceiling scores, could you share how prize-winning submissions are verified (e.g. code/model review, reproduction on held-out data)?
Thanks!
Same question is being tracked on the challenge page, with some extra sub-questions: https://huggingface.co/spaces/nvidia/PhysicalAI-AV-OOD-Reasoning-Challenge-2026/discussions/8. A ruling there would be much appreciated.