search

argoverse / av2-api

Argoverse 2: Next generation datasets for self-driving perception and forecasting.
https://argoverse.github.io/user-guide/
MIT License
306 stars 71 forks source link

Motion prediction submitssion #229

Closed soyons closed 8 months ago

soyons commented 8 months ago

I am newer to the challenge and have the following questions when I want to submit the forecasting result. Please kindly to help me.

  1. Where can I find the submit data format in eval.ai as in can not see any detail instructions argoverse.github.io.
  2. I want to participate the muti-agent motion forecasting challenge but I don't know which timestamp and agent should I concern, can anybody tell me the detail instructions?

Many thanks.

soyons commented 8 months ago

find solution

cancui19 commented 8 months ago

@soyons Hi, would you mind kindly letting me know how you solve this issue? I cannot find the instructions either...

soyons commented 8 months ago

@soyons Hi, would you mind kindly letting me know how you solve this issue? I cannot find the instructions either... I found the result from https://github.com/jchengai/forecast-mae/blob/main/src/model/multiagent/model_forecast_ma.py#L81 It may give you the answer by reading the above codes.

In av2 dataset, the past 5s frames are feed into encoder and we predict the next 6s frames. So the the front 50 frame is set to be visible in parquet files. In multi agent prediction task, only one train sample can be generate per parquet file but of course you can use the data argumentation technique.

cancui19 commented 8 months ago

@soyons Thanks for following up. One more question is that we cannot access the submission instructions (argoverse.github.io) for eval.ai. Could you show us the instructions for submitting our results? Thanks in advance.