search

Tsinghua-MARS-Lab / M2I

M2I is a simple but effective joint motion prediction framework through marginal and conditional predictions by exploiting the factorized relations between interacting agents.
https://tsinghua-mars-lab.github.io/M2I/
MIT License
187 stars 24 forks source link

difference between 'influencer' and 'dominant' #11

Closed yanwu-ge closed 1 year ago

yanwu-ge commented 1 year ago

In src/dataset_waymo.py, line 290, there is some comments as follows: ' # [id1, id2, label, relation_type].' ' # id1 always influencer, id2 always reactor' ' # interaction_label: 1 - bigger id agent dominant, 0 - smaller id agent dominant, 2 - no relation' ' # agent_pair_label: 1 - v2v, 2 - v2p, 3 - v2c, 4 - others'

I am confused about 'influencer' and 'dominant'. In my understanding, 'influencer' dominates the interaction. So why are there there cases in 'interaction_label' ? Besides, when will the code used for generating the ground truth label released ? Looking forward to your reply.

larksq commented 1 year ago

The interaction label which is a value in [0, 1, 2] is a redundancy designed for the ground truth label for the classification training. Examples are as follows: [100, 0, 1, 1] or [0, 100, 0, 1]

As for the codes used for generating the ground truth labels, sadly we don't have a plan to release them yet. These codes are frequently updated in our later projects and will not be compatible with earlier ones. A basic generator we used in this paper is nothing more than a series of bounding boxes overlap checking. Shoot me an email if you need more than a toy-level label generator.