search

sshaoshuai / MTR

MTR: Motion Transformer with Global Intention Localization and Local Movement Refinement, NeurIPS 2022.
Apache License 2.0
688 stars 112 forks source link

No idea why this happens, as previously I didn't notice the visualization carefully. #45

Closed ggosjw closed 1 year ago

ggosjw commented 1 year ago 
          No idea why this happens, as previously I didn't notice the visualization carefully.

Hope to see whether other people also have these unnormal effects.

Originally posted by @sshaoshuai in https://github.com/sshaoshuai/MTR/issues/39#issuecomment-1575495051

ggosjw commented 1 year ago

Thank you for making your fantastic prediction framework available to the public. While visualizing the MTR prediction outcomes, I encountered a scenario that resembles the aforementioned quoting issue. Initially, the predictions are accurate in the first frame, but they become less stable over time, culminating in a circular trajectory prediction towards the end (as shown in the video).

https://github.com/sshaoshuai/MTR/assets/87901530/f2ca19f4-edba-4285-aef8-eb95ccd37443

Could you possibly offer some guidance or insights into this unusual outcome? Any wisdom or assistance you can provide would be greatly appreciated. Thank you.

jiaxiaosong1002 commented 1 year ago

Hi ggosjw,

Your visualization code is great! Could you please provide this snippet so that I could check my output as well? Thanks!

ggosjw commented 1 year ago

Hi ggosjw,

Your visualization code is great! Could you please provide this snippet so that I could check my output as well? Thanks!

I appreciate your suggestions, but I don't intend to share any visualization code at the moment due to personal reasons. Thank you for your understanding. It primarily involves some work with pyplot.