uber-research / LaneGCN

[ECCV2020 Oral] Learning Lane Graph Representations for Motion Forecasting
https://arxiv.org/abs/2007.13732
Other
496 stars 131 forks source link

confused about the 'theta' param in actor preprocessing #13

Closed johnnyleeRH closed 3 years ago

johnnyleeRH commented 3 years ago

HI @chenyuntc, thans for your source code, I'm confused about the theta param in data:get_obj_feats recently. I usually get theta from actan2( (pos[19].y - pos[18].y) , (pos[19].x - pos[18].x)). what's your calculation specital mean here?

theta = np.pi - np.arctan2(pos[18].y-pos[19].y, pos[18].x - pos[19].x)

BR~