FTGN_Weights

Trained model checkpoint with the hyper-parameters:

--feature_dim 128
--batch_size 64
--epoch 36
--learning_rate 1e-4
--weight_decay 0
--lr_decay_schedule 0
--pretrained_vectornet
--time_split_size 2
--fuse_time
--sequential_encoding
--scene_memory
--endpoint_source both
--temperature 0.3
--temperature_validation
--lambda1 1.0
--lambda2 1.0
--lambda3 1.0

Here is the detailed evaluation results:

minADE: 0.73048
minFDE: 1.07642
MR: 0.10471
p-minADE: 2.43828
p-minFDE: 2.78422
p-MR: 0.77483
brier-minADE: 1.33877
brier-minFDE: 1.68472