search

Henry1iu / TNT-Trajectory-Prediction

A Pytorch Implementation of TNT: Target-driveN Trajectory Prediction
487 stars 92 forks source link

fully connected subgraphs? #15

Closed linaashaji closed 2 years ago

linaashaji commented 2 years ago

Hello,

During constructing edge indices (get_fc_edge_index --> _get_x method --> argoverse_loader_v2.py), I realized that you are building a fully connected graph for each cluster. However, as far as I understand your code, this cluster is constructed at the sub-graph level. So, shouldn't you connect the nodes according to the topology of the sub-graph? (e.g., the connections for the crosswalk are different from the connections in a regular lane).

Thank you, Lina

Henry1iu commented 2 years ago

Hi,

You can refer to Section 3.2 of the (VectorNet paper). They constructed a fully connected graph at the subgraph level.

Your suggestion sounds more reasonable. However, I did try to construct the subgraph according to the topology order. It showed worse performance. If you're interested, you can verify it yourself. If you find anything interesting, please do share it with me. :)

Best Regards, Jianbang

linaashaji commented 2 years ago

Okay, thank you!