Some confusions about this code

[jackson-sunj]

Hi,
Thank you for sharing the great work. I have some confusions about the code, and I put these questions in a PDF file, please check it out, thanks for your reply！
Some confusions about this code.pdf

[GentleSmile]

1. Loss for lane can be found in 'lane_scoring' module (in our updated code)
2. New detailed description for this function may be helpful
3. Scores of heatmap indicate probability distribution of final position. Generated 6 trajectories together with their scores can be treated as a simplification of the heatmap.

[jackson-sunj]

Thank for your reply!
I have another question about Goal_2D, in the code, the function def get_subdivide_points(polygon): is still dense sampling on the lane centerline, where can it be expressed that it is dense sampling on the lane?

[GentleSmile]

We treat sampling on the centerline as sparse sampling. Dense sampling is performed at

DenseTNT/src/modeling/decoder.py

Line 143 in a0e3b8a

goals_2D_dense = utils.get_neighbour_points(goals_2D[topk_ids], topk_ids=topk_ids, mapping=mapping[i])

[jackson-sunj]

Thank for your reply!
I also want to ask you a question，without using " a set predictor ", we calculate the loss in " goals_2D_per_example_calc_loss ",we get a " predict_traj " It's just to calculate loss? and we get 6 trajectories using the topk(6) scores. How do we determine that the calculated " predict_traj " must be in the 6 trajectories with the highest score in the end?

[weihuaLi1]

I also have the similar question. We get the "highest_goal" in

DenseTNT/src/modeling/decoder.py

Line 228 in a0e3b8a

scores, highest_goal, goals_2D = \

But it not used for generating the trajectory. So what is the usage of it?

[GentleSmile]

predict_traj here is only used to calculate loss for trajectory completion. During training, we don't need to generate top K goals. During evaluation, we get top K goals at

DenseTNT/src/modeling/decoder.py

Line 263 in a0e3b8a

pred_goals_batch, pred_probs_batch = utils.select_goals_by_optimization(