A single RGB image plus a language instruction predicts a metric 6-DoF object trajectory, decoded from the intermediate latents of a frozen video diffusion model at an early denoising step.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DreamTraj: Generating 6-DoF Object Trajectories by Reading Unrendered Video Diffusion Latents
A single RGB image plus a language instruction predicts a metric 6-DoF object trajectory, decoded from the intermediate latents of a frozen video diffusion model at an early denoising step.