Dyn failure audit
15 valid, scene-deduplicated low-scoring cases from the new eval.
Open visualSTRIVE · 08/28/2026
Four small, auditable views of the Dyn generalization, prefill, and mixed-data experiments.
15 valid, scene-deduplicated low-scoring cases from the new eval.
Open visual15 shared clips across Baselines 1, 2, 3, and E2E.
Open visual10 standard MOVi-E and 10 Dyn-adapted MOVi-E previews.
Open visual10 examples from each of val/test × 24/120 frames.
Open visual