End-to-End-Multi-View-Fusion-for-3D-Object-Detection-in-LiDAR-Point-Clouds icon indicating copy to clipboard operation
End-to-End-Multi-View-Fusion-for-3D-Object-Detection-in-LiDAR-Point-Clouds copied to clipboard

poor performance

Open CBY-9527 opened this issue 4 years ago • 3 comments

Thank you very much for your code. I am interested in MVF, but the performance of the code you provide is very poor. The performance of 3D is almost 0. Can you tell me the possible reasons? 2020-12-11 09:11:02,546 INFO Car [email protected], 0.70, 0.70: bbox AP:0.0000, 0.0017, 0.0068 bev AP:55.6617, 49.9037, 49.5705 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00 Car [email protected], 0.70, 0.70: bbox AP:0.0000, 0.0005, 0.0019 bev AP:54.2903, 48.2406, 47.3495 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00 Car [email protected], 0.50, 0.50: bbox AP:0.0000, 0.0017, 0.0068 bev AP:73.4910, 69.5406, 70.6138 3d AP:0.0038, 0.0129, 0.0176 aos AP:0.00, 0.00, 0.00 Car [email protected], 0.50, 0.50: bbox AP:0.0000, 0.0005, 0.0019 bev AP:74.4895, 69.5168, 69.9116 3d AP:0.0010, 0.0035, 0.0048 aos AP:0.00, 0.00, 0.00

CBY-9527 avatar Dec 11 '20 01:12 CBY-9527

first evaluate the model I provide, I don't know why you get the poor performance, maybe you should use the same setting.

AndyYuan96 avatar Dec 31 '20 05:12 AndyYuan96

first evaluate the model I provide, I don't know why you get the poor performance, maybe you should use the same setting.

I also got poor results if I trained the model from scratch with the default setting.

mvf_pp_dv: Car [email protected], 0.70, 0.70: bbox AP:0.0008, 0.0019, 0.0038 bev AP:62.8055, 54.6709, 54.6212 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00

mvf_paper_car: Car [email protected], 0.70, 0.70: bbox AP:0.0000, 0.0013, 0.0073 bev AP:55.4981, 49.7647, 49.4458 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00

But just testing with your provided pth file is OK --- the results are similar. I also wonder the possible reasons.

yytzsy avatar Aug 02 '21 12:08 yytzsy

first evaluate the model I provide, I don't know why you get the poor performance, maybe you should use the same setting.

I also got poor results if I trained the model from scratch with the default setting.

mvf_pp_dv: Car [email protected], 0.70, 0.70: bbox AP:0.0008, 0.0019, 0.0038 bev AP:62.8055, 54.6709, 54.6212 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00

mvf_paper_car: Car [email protected], 0.70, 0.70: bbox AP:0.0000, 0.0013, 0.0073 bev AP:55.4981, 49.7647, 49.4458 3d AP:0.0000, 0.0000, 0.0000 aos AP:0.00, 0.00, 0.00

But just testing with your provided pth file is OK --- the results are similar. I also wonder the possible reasons.

I meet the same problem like yours, have you solved it ?

wanghangege avatar Sep 02 '23 13:09 wanghangege