Category-agnostic recall under continuous motion and occlusion.
NSTM is prior-free and generalizes beyond humans as shown by our Category-Agnostic Object-Centric NVS examples below. Although synthetic, this dataset remains highly challenging due to its complex occlusion patterns. Significant ambiguity arises in novel view synthesis when relying on a single view or partial observations, which manifests as jittering in our stateless baseline (LVSM).
Select a scene, then click models in the comparison strip to inspect them in detail below. Check the memory box to see our memory readout visuals.