The amount of data generated by prepare_data_dwn is not the same

[stanny880913]

Hello, I trained the structure of your paper twice and ran all the processes completely. However, when I did the --do_eval of train_dwn again, I found that the amount of data was different. I used the ones generated through prepare_data_dwn. Val.json is used for verification. Is this normal? Or could something possibly go wrong? thank you~

[longyunf]

You can check if the arguments of dataloader are the same. https://github.com/longyunf/radiant/blob/cf5355396d42ef17940e29ef8f9e3cabfd8035c3/scripts/train_dwn.py#L286-L289

[stanny880913]

You can check if the arguments of dataloader are the same.

https://github.com/longyunf/radiant/blob/cf5355396d42ef17940e29ef8f9e3cabfd8035c3/scripts/train_dwn.py#L286-L289

Hello, my arguments of dataloader are all the same, Will the model accuracy of radar branch affect the amount of data? thank you

[longyunf]

The accuracy of radar branch may influence radar-camera association results.

[longyunf]

You can also check if val.json changes.

[stanny880913]

The accuracy of radar branch may influence radar-camera association results.

Hello, so the higher the accuracy of the radar branch model, the more data will be generated by prepare_dwn_data(include train.json and val.json)?

[stanny880913]

The accuracy of radar branch may influence radar-camera association results.

How can I check radar branch model accuracy, Is it output through error_rd of --do_eval in train_dwn.py? Determine the quality of the radar branch model based on the loss value? thank you