You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
To prevent confusion of the return type in biogtr.models.Transformer, the only output of the transformer should be the association matrix. If we want to store the position embeddings for debugging we should instead save it at the frame or instance level (depending on what level of granularity we want). Instance level is probably better considering we also store instance features/crops/etc at the instance level and we want to move from using Frame objects as inputs to the model to Instance Objects. Thus, we need to add a embedding attribute to the Instance object similar to how we store the trajectory scores.
The text was updated successfully, but these errors were encountered:
To prevent confusion of the return type in
biogtr.models.Transformer
, the only output of the transformer should be the association matrix. If we want to store the position embeddings for debugging we should instead save it at the frame or instance level (depending on what level of granularity we want). Instance level is probably better considering we also store instance features/crops/etc at the instance level and we want to move from usingFrame
objects as inputs to the model toInstance
Objects. Thus, we need to add aembedding
attribute to theInstance
object similar to how we store the trajectory scores.The text was updated successfully, but these errors were encountered: