Open pjw1 opened 2 years ago
Hi! Thank you for kindly sharing this code.
I wonder if I can reproduce the results (ADE score) in SimNet paper. The paper says the ADE is 0.49, 0.76, 1.32 at 0.5, 1, 2sec respectively.
But the pre-trained simulation model can't reproduce the score and the gap is quite huge. (https://lyft-l5-datasets-public.s3-us-west-2.amazonaws.com/models/simulation_models/simulation_model_20210416_5steps.pt) 0.96, 1.42, 1.64 at 0.5, 1, 2.
Can you share the model used in the paper or implementation details to reproduce the result?
Thank you in advance!
Hi! Thank you for kindly sharing this code.
I wonder if I can reproduce the results (ADE score) in SimNet paper. The paper says the ADE is 0.49, 0.76, 1.32 at 0.5, 1, 2sec respectively.![image](https://user-images.githubusercontent.com/19548825/143677533-11fae0d5-b328-44db-818a-eefbdd033568.png)
But the pre-trained simulation model can't reproduce the score and the gap is quite huge. (https://lyft-l5-datasets-public.s3-us-west-2.amazonaws.com/models/simulation_models/simulation_model_20210416_5steps.pt) 0.96, 1.42, 1.64 at 0.5, 1, 2.
Can you share the model used in the paper or implementation details to reproduce the result?
Thank you in advance!