Why use ''mashed_scatter_" before conv social pooling in the model?
Closed this issue · 1 comments
RomanticLeon commented
- Why use ''mashed_scatter_" before conv social pooling in the model? How to get mash from data?
- The neighbor's pass position data is converted a tensor('nbrs') input to the model? But in the paper, every neighbor's position is inputted a corresponding LSTM.
nachiket92 commented
masked_scatter allows for more efficient batching for the encoder LSTMs and the conv social pooling layers during training. There will typically be a different number of neighboring vehicles for each training instance. During the forward pass, all neighboring vehicles in a minibatch are pooled together and processed by the encoder LSTM. The masked_scatter function then places the LSTM states for each vehicle at the appropriate cell location, for the appropriate training instance.