Closed susanbao closed 2 years ago
Hi, this is easy to achieve with minimum extra coding. Use late fusion strategy but don't include any other cav's results in the late fusion stage.
Just an alternative approach: I implemented no fusion by removing all the LiDAR pcd from other CAV but keep the groundtruth there.
Hi, this is easy to achieve with minimum extra coding. Use late fusion strategy but don't include any other cav's results in the late fusion stage.
Would it be correct if I get the test results of no fusion by using the training strategy of late fusion and without including any other cav's results in the inference_late_fusion function?
yes, that is correct, although myself separately train the models for no fusion (I didn't use the already-trained late fusion model directly). But you can do the thing you said also if you want to save some time.
yes, that is correct, although myself separately train the models for no fusion (I didn't use the already-trained late fusion model directly). But you can do the thing you said also if you want to save some time.
Thanks for your help.
Hi, this is easy to achieve with minimum extra coding. Use late fusion strategy but don't include any other cav's results in the late fusion stage.
Hello,firstly thank you for the work you do! Excuse me, what should I do to achieve this? Can I achieve this result if I delete the CAV folder under each scenario folder until there is only one left?
Hello, in your paper, there are experiment results on no fusion part. But there is no corresponding yaml files on no fusion. How do you get the results? Train and test with OpenPCDet on opv2v data?