MDST-DGCN: A Multilevel Dynamic Spatiotemporal Directed Graph Convolutional Network for Pedestrian Trajectory Prediction

Table 1

We compare deterministic baseline models with MDST-DGCN of deterministic type (MDST-DGCN-D) and stochastic baseline models with MDST-DGCN of stochastic type (MDST-DGCN-S) on ETH and UCY.

Method	ETH	HOTEL	UNIV	ZARA1	ZARA2	AVG
LSTM [14, 27]	1.09/2.41	0.86/1.91	0.61/1.31	0.41/0.88	0.52/1.11	0.70/1.52
S-LSTM [11, 14]	1.09/2.35	0.79/1.76	0.67/1.40	0.47/1.00	0.56/1.17	0.72/1.54
Social Attention [15, 22]	1.39/2.39	2.51/2.91	1.25/2.54	1.01/2.17	0.88/1.75	1.41/2.35
CIDNN [19, 22]	1.25/2.32	1.31/2.36	0.90/1.86	0.50/1.04	0.51/1.07	0.89/1.73
MDST-DGCN-D	0.86/1.75	0.44/0.90	0.55/1.16	0.40/0.86	0.32/0.68	0.51/1.07
SoPhie [16]	0.70/1.43	0.76/1.67	0.54/1.24	0.30/0.63	0.38/0.78	0.54/1.15
GAT [21]	0.68/1.29	0.68/1.40	0.57/1.29	0.29/0.60	0.37/0.75	0.52/1.07
Social-BiGAT [21]	0.69/1.29	0.49/1.01	0.55/1.32	0.30/0.62	0.36/0.75	0.48/1.00
SGAN [14]	0.81/1.52	0.72/1.61	0.60/1.26	0.34/0.69	0.42/0.84	0.58/1.18
SGAN-p [14]	0.87/1.62	0.67/1.37	0.76/1.52	0.35/0.68	0.42/0.84	0.61/1.21
STGAT [22]	0.65/1.12	0.35/0.66	0.52/1.10	0.34/0.69	0.29/0.60	0.43/0.83
MDST-DGCN-S	0.69/1.45	0.34/0.58	0.51/1.11	0.33/0.70	0.28/0.59	0.43/0.89
Social-STGCNN [23]	0.64/1.11	0.49/0.85	0.44/0.79	0.34/0.53	0.30/0.48	0.44/0.75
MDST-DGCN-S3	0.56/1.12	0.27/0.50	0.38/0.81	0.27/0.56	0.22/0.46	0.34/0.69

We predict future at 4.8 seconds (short-term prediction), given the previous 3.2 seconds. The errors reported are ADE or FDE in meters. Methods marked with draw 20 samples. The ADE and FDE of methods marked with superscript 2 are calculated by selecting the closest sample; the ADE and FDE of methods marked with superscript 3 are calculated by selecting the closest trajectory; and for the ADE and FDE of methods marked with superscript 1, we are not sure which type they belong to, because we cannot find their code. The values with the least error in the comparison model are bolded.