Why model output pred_score is all zeros?

[evercherish]

when inferencing set predictor model, why the output pred_score is all zerors?
I notice that pred_probs is initialized with 'np.zeros((batch_size, 6))' in function: goals_2D_eval() if 'set_predict', and as a output directly.
So we dont need a pred_probs in set predictor mode?
or I miss something?

[GentleSmile]

When we were developing the code, only Argoverse 1 was available. And the evaluation in Argoverse 1 did not require outputting scores (minFDE, Miss Rate), so the implementation of the set predictor did not consider outputting scores.

[evercherish]

thx for your reply!
Another question, seeking an answer plz.
If I wanna add a score for each pred trajection, should I modify output shape of module complete_traj_decoder from (6,60,2) to (6,60,3)? The extra dimention is score of each point (loss is defined by distance of pred point and label point), and the trajection score is obtained by multiplying these scores.
Or simply use the goal‘s score to represent the trajectory's score?

[GentleSmile]

Maybe multiplying these scores to represent a trajectory's score is not reasonable. It would be better to train a score for each trajectory, treating it as a 6-class classification instead of 60-class classification. If using the goal‘s score to represent the trajectory's score, some normalization is required.

[evercherish]

thx!!!