changyanchuan / trajcl Goto Github PK

Hi,
I have a question about using your proposed method on non-GPS datasets, i.e., the Opentraj dataset which is the (x,y) pixel of moving pedestrians. I understand that I need to come up with a new preprocessing file for this new dataset. I'm wondering if there is a specific point or guideline that I can follow for this experiment.

Could you please inform me whether you selected the L2 Euclidean distance or the cosine distance when calculating the trajectory embedding vector similarity using faiss? I look forward to your reply.

Hi, thank you for the great contribution.

I wonder why you set 200000 samples as the training set instead of the full set (above 900000 samples).

Thank you.

Hello, I am studying your article "trajcl", and now I am encountering some environmental issues. My graphics card is A10, the base image is pytorch:1.14-cu12.0-py3.8-ubuntu20.04. Now I have created an environment consistent with your version using conda env: Ubuntu 20.04+Python 3.7.7+torch-1.8.1+cu102, but now when I import torch, I encounter a coredump, and have not found a way to fix it yet. I am thinking of using the higher version "pytorch:1.14-cu12.0-py3.8-ubuntu20.04" to run your model, could this work? xiexie for u reply. lol