What's the meaning of v2xset_4modality.json?

[deyang2000]

Hello, yifan. I trained directly with the original v2xset data set, and everything worked fine. Now, I have made some modifications to the original data set v2xset. Specifically, I have redistributed the files in the original training, verification, and test sets to get a new data set. (Equivalent to only changing the file path) There are some errors when I train with the new data set, which is related to the v2xset_4modality.json file.

I noticed that in the camera pyramid.yaml file ego is set to m2, which should be the error caused here. What rules are used to assign m1/m2/m3/m4 in the v2xset_4modality.json file? How do I make changes based on my new data set? I would appreciate your prompt reply.

[deyang2000]

Specifically, my new dataset looks like this: Total scenes: 58 29 contain roadside facilities. The data set is divided into: Train_vehicle:training dataset for vehicle .(29) Val_vehicle:validate dataset for vehicle .(29) New_scene: New scenes including V2I,which is the same as Val_vehicle.(29)

[deyang2000]

I have identified the problem and am now training the model, hoping that this new data set will be helpful for my further research.

[yifanlu0227]

I am glad the problem has been solved.

• The identifiers like m1, m2 in v2xset_4modality.json are used to assign agent type to each agent in the scene. With this assignment, we ensure the validation scenarios for all methods are consistent and fixed.

• The identifiers like m1, m2 in ${METHOD}.yaml are used to specify the sensor configuration and detection model used by this agent type (like m2 in the case of camera_pyramid.yaml).

In ${METHOD}.yaml, there is also a concept of mapping_dict. It maps the given agent type from v2xset_4modality.json to another agent type in the experiment. As you can see, camera_pyramid.yaml is a homogeneous collaborative perception setting, so the type of all agents should be the same, which can be referred to by m2.

Just note that mapping_dict will not take effect during the training process to introduce more data augmentation. Each agent will be randomly assigned an agent type that exists in yaml.

[deyang2000]

Thank you very much for your answer! So can I take it this way: There is a folder named "2021_08_18_21_38_28" in v2x_train, and the subfolders of "2021_08_18_21_38_28" are not assigned to m2 in V2Xset_4modalm. json. So "2021_08_18_21_38_28" is not used as training data actually.

"2021_08_18_21_38_28": {
    "8786": "m3",
    "8795": "m4"
}

[yifanlu0227]

No. If the mapping_dict is

  mapping_dict:
    m1: m2
    m2: m2
    m3: m2
    m4: m2

the "8786" and "8795" agents will participate in the training as m2 agent type.

[deyang2000]

Oooooh, I misunderstood. Thank you.