About the question of velocity normlization

[zJJjing]

Thank you very much for sharing. My recent research involves the social implicit model, but I have encountered an issue as shown in the figure. The input v. permute (0, 3, 1, 2) [0,:,:, 0] vectors for the norm operation are all 0, not the maximum value for each trajectory. I'm not sure if my understanding was incorrect. Looking forward to your answer, thank you！