Open mohamedsamirx opened 10 months ago
We have not tried this combination. Since NMP focuses on improving BEV feature representation with historical cached feature, which is orthogonal to MapTRv2. We will consider adding this feature in the future.
okay that is gonna be awesome , like do you have any ideas on how we can do it? my idea was to add the c2p attention and gru block with the map and the prior feature and do the global update on it with using the maptrv2 decoder as the same and add the global map encoder to the encoder block. correct me if i am wrong .
is it possible to merge neural map prior with maptrv2 model i have tried to do it but i couldn't ?