DeltaFlow_RVloss_AV2

IV ·2026-05-01 unsupervised single-dataset archived paper / code ↗

mean Dynamic · 0–35 m

0.2720

mean Static · 0–35 m

0.0205

Threeway EPE · 0–35 m

0.1248

Notes

Backbone: DeltaFlow Loss: RunoffVoteLoss (self-supervised) Training: Argoverse 2 only; no additional data; Paper Link: https://arxiv.org/abs/2608.18864

Distance range 0-35 m

Metric	Aeva	AV2	nuScenes	TruckScenes	Waymo	Overall
mean Dynamic	0.3273	0.1632	0.3702	0.3183	0.1813	0.2720
mean Static	0.0233	0.0164	0.0124	0.0371	0.0132	0.0205
Threeway EPE	0.1755	0.0295	0.0351	0.3420	0.0420	0.1248
Threeway FG Dynamic	0.4573	0.0492	0.0708	0.8978	0.0981	0.3146
Threeway FG Static	0.0330	0.0212	0.0175	0.0114	0.0115	0.0189
Threeway BG Static	0.0361	0.0182	0.0171	0.1168	0.0164	0.0409
CAR Dynamic	0.1946	0.1161	0.2377	0.1179	0.1111	0.1555
CAR Static	0.0182	0.0184	0.0103	0.0188	0.0086	0.0149
OTHER_VEHICLES Dynamic	0.3952	0.1541	0.3695	0.3322	—	0.3128
OTHER_VEHICLES Static	0.0188	0.0172	0.0112	0.0135	—	0.0152
PEDESTRIAN Dynamic	0.3380	0.2421	0.5619	0.4951	0.2780	0.3830
PEDESTRIAN Static	0.0233	0.0130	0.0135	0.0158	0.0110	0.0153
WHEELED_VRU Dynamic	0.3812	0.1406	0.3115	0.3280	0.1548	0.2632
WHEELED_VRU Static	0.0223	0.0161	0.0104	0.0209	0.0186	0.0177
BACKGROUND Static	0.0337	0.0170	0.0169	0.1165	0.0144	0.0397

Not evaluated at 35–70 m. The team chose to evaluate 0–35 m only (e.g. the model crops the point cloud at 51.2 m), so this entry is ranked at 0–35 m only.

Imported from the original leaderboard at sceneflow.argoverse.org.