Results on SDD

[DelinquentLeon]

Thanks for your excellent work! I wondered if there are results on dataset SDD. If not, how could i evaluate on SDD.

[InhwanBae]

Hi @DelinquentLeon,

Thanks for your interest in our work!

Of course, we have results on the Stanford Drone Dataset (SDD).

Stanford Drone Dataset (SDD)	PECNet	MID	SingularTrajectory
ADE/FDE	9.97 / 15.89	7.60 / 14.37	7.58 / 12.06

Grand Central Station (GCS) dataset	PECNet	MID	SingularTrajectory
ADE/FDE	17.08 / 29.30	10.72 / 18.20	7.12 / 11.98

nuScenes	Metric	CoverNet	DSF-AF	Trajectron++	AgentFormer	SingularTrajectory
Samples = 5	ADE₅	1.96	2.06	1.88	1.86	1.42
	FDE₅	-	4.67	-	3.89	3.01
Samples = 10	ADE₁₀	1.48	1.66	1.51	1.45	1.17
	FDE₁₀	-	3.71	-	2.86	2.33

Additionally, if needed, you can use the following transferability table, which is crucial for the universal trajectory predictor.

Cross-Dataset Evaluation

Train → Test	PECNet	MID	SingularTrajectory
SDD → SDD	9.97 / 15.89	7.60 / 14.37	7.58 / 12.06
ETH/UCY → SDD	11.42 / 19.45	11.74 / 21.95	7.82 / 12.61
Gain ↑	-14.5% / -22.4%	-54.5% / -52.7%	-3.2% / -4.6%

Train → Test	PECNet	MID	SingularTrajectory
GCS → GCS	17.08 / 29.30	10.72 / 18.20	7.12 / 11.98
ETH/UCY → GCS	23.08 / 40.50	18.53 / 31.86	7.71 / 13.21
Gain ↑	-35.1% / -38.2%	-72.9% / -75.1%	-8.3% / -10.3%

Cross-Task Evaluation

Train \ Test	Stochastic	Few-shot	Momentary
Stochastic	0.21 / 0.32	0.21 / 0.32	0.26 / 0.42
Few-shot	0.23 / 0.36	0.23 / 0.35	0.27 / 0.44
Momentary	0.23 / 0.37	0.23 / 0.37	0.25 / 0.40