search

narsisn / Argoverse2_Motion_Forecasting

MFTF: Motion Forecasting Using Transformers
Other
50 stars 4 forks source link

Documentation of this repo #2

Open Cram3r95 opened 1 year ago

Cram3r95 commented 1 year ago

Hi,

Thank you for your wonderful work. Could you provide more details about the structure of the pipeline? What are the differences between TGR and MTMF models?

Comparing TGR and Crat-PRED, you replace the LSTM encoder with a Transformer encoder (Linear encoding + Positional Encoding + Transformer Encoder). Then, the interaction graph (Crystal-GCN) and Linear residual is similar to Crat-Pred. What about MTMF? Any paper or documentation?

Congrats for your work!

narsisn commented 1 year ago

Hi, Thank you for your interest. Could you please provide me your email address, so I will send you some documents.

Cram3r95 commented 1 year ago

Sure, my email is: carlos.gomezh@uah.es

I am trying to run the code with cuda 11.6 and the corresponding pytorch version, though you have other pytorch and specially torch-geometric versions. It is a headache when you try to integrate the different dependencies using conda + docker haha.