Open Ruoyu-Xu opened 2 years ago
ylwhxht
commented
1 year ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
Ruoyu-Xu
commented
1 year ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
ylwhxht
commented
1 year ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
Thank you. Is it in AP_R40@0.5 0.5 0.5?
My reproduction result is like this, and it looks the same
Car AP@0.70 0.70, 0.70:
Bbox AP: 36.8576, 32.3523, 31.4710
Bev AP: 38.7652, 35.5619, 34.8579
3D AP: 17.2034, 15.0750, 15.0462
AOS AP: 36.46, 32.01, 31.09
Car AP_R40@0.70 0.70, 0.70:
Bbox AP: 34.1645, 29.5950, 28.4427
Bev AP: 36.4198, 32.3735, 31.4788
3D AP: 14.3970, 12.3787, 11.8986
AOS AP: 33.77, 29.22, 28.09
Car AP@0.50 0.50, 0.50:
Bbox AP: 59.9559, 52.9196, 51.8491
Bev AP: 59.0141, 51.9162, 51.0499
3D AP: 49.5782, 42.9737, 42.7284
AOS AP: 59.16, 52.15, 50.94
Car AP_R40@0.50 0.50, 0.50:
Bbox AP: 59.3984, 51.7946, 50.4342
Bev AP: 59.0424, 50.9988, 48.9314
3D AP: 48.2886, 41.6285, 39.8300
AOS AP: 58.52, 50.89, 49.54
Ruoyu-Xu
commented
1 year ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
Thank you. Is it in AP_R40@0.5 0.5 0.5?
My reproduction result is like this, and it looks the same
Car AP@0.70 0.70, 0.70:
Bbox AP: 36.8576, 32.3523, 31.4710
Bev AP: 38.7652, 35.5619, 34.8579
3D AP: 17.2034, 15.0750, 15.0462
AOS AP: 36.46, 32.01, 31.09
Car AP_R40@0.70 0.70, 0.70:
Bbox AP: 34.1645, 29.5950, 28.4427
Bev AP: 36.4198, 32.3735, 31.4788
3D AP: 14.3970, 12.3787, 11.8986
AOS AP: 33.77, 29.22, 28.09
Car AP@0.50 0.50, 0.50:
Bbox AP: 59.9559, 52.9196, 51.8491
Bev AP: 59.0141, 51.9162, 51.0499
3D AP: 49.5782, 42.9737, 42.7284
AOS AP: 59.16, 52.15, 50.94
Car AP_R40@0.50 0.50, 0.50:
Bbox AP: 59.3984, 51.7946, 50.4342
Bev AP: 59.0424, 50.9988, 48.9314
3D AP: 48.2886, 41.6285, 39.8300
AOS AP: 58.52, 50.89, 49.54
Well, maybe there are some bugs in my code. I did it a year ago and I forgot many details.
ylwhxht
commented
1 year ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
Thank you. Is it in AP_R40@0.5 0.5 0.5? My reproduction result is like this, and it looks the same Car AP@0.70 0.70, 0.70: Bbox AP: 36.8576, 32.3523, 31.4710 Bev AP: 38.7652, 35.5619, 34.8579 3D AP: 17.2034, 15.0750, 15.0462 AOS AP: 36.46, 32.01, 31.09 Car AP_R40@0.70 0.70, 0.70: Bbox AP: 34.1645, 29.5950, 28.4427 Bev AP: 36.4198, 32.3735, 31.4788 3D AP: 14.3970, 12.3787, 11.8986 AOS AP: 33.77, 29.22, 28.09 Car AP@0.50 0.50, 0.50: Bbox AP: 59.9559, 52.9196, 51.8491 Bev AP: 59.0141, 51.9162, 51.0499 3D AP: 49.5782, 42.9737, 42.7284 AOS AP: 59.16, 52.15, 50.94 Car AP_R40@0.50 0.50, 0.50: Bbox AP: 59.3984, 51.7946, 50.4342 Bev AP: 59.0424, 50.9988, 48.9314 3D AP: 48.2886, 41.6285, 39.8300 AOS AP: 58.52, 50.89, 49.54
Well, maybe there are some bugs in my code. I did it a year ago and I forgot many details.
Okay, thank you very much. After I reproduced it, I think there might be some issues with my code, too.
But in fact, I almost haven't changed the code, so I don't know how to solve it. I see that you have the same problem with the same performance, so I asked if you have solved it.
c-yyyy
commented
4 months ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
Thank you. Is it in AP_R40@0.5 0.5 0.5? My reproduction result is like this, and it looks the same Car AP@0.70 0.70, 0.70: Bbox AP: 36.8576, 32.3523, 31.4710 Bev AP: 38.7652, 35.5619, 34.8579 3D AP: 17.2034, 15.0750, 15.0462 AOS AP: 36.46, 32.01, 31.09 Car AP_R40@0.70 0.70, 0.70: Bbox AP: 34.1645, 29.5950, 28.4427 Bev AP: 36.4198, 32.3735, 31.4788 3D AP: 14.3970, 12.3787, 11.8986 AOS AP: 33.77, 29.22, 28.09 Car AP@0.50 0.50, 0.50: Bbox AP: 59.9559, 52.9196, 51.8491 Bev AP: 59.0141, 51.9162, 51.0499 3D AP: 49.5782, 42.9737, 42.7284 AOS AP: 59.16, 52.15, 50.94 Car AP_R40@0.50 0.50, 0.50: Bbox AP: 59.3984, 51.7946, 50.4342 Bev AP: 59.0424, 50.9988, 48.9314 3D AP: 48.2886, 41.6285, 39.8300 AOS AP: 58.52, 50.89, 49.54
Well, maybe there are some bugs in my code. I did it a year ago and I forgot many details.
Okay, thank you very much. After I reproduced it, I think there might be some issues with my code, too.
But in fact, I almost haven't changed the code, so I don't know how to solve it. I see that you have the same problem with the same performance, so I asked if you have solved it.
hello,so have you solved the dataset split question？
ylwhxht
commented
4 months ago Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.
Hello, I also tried to randomly partition the dataset, but the accuracy after reproduction is not good. Do you remember the performance results of reproduction?
For the baseline PointPillars with only LiDAR input, I got about 8 percent drop in 3D view at moderate level.
Thank you. Is it in AP_R40@0.5 0.5 0.5? My reproduction result is like this, and it looks the same Car AP@0.70 0.70, 0.70: Bbox AP: 36.8576, 32.3523, 31.4710 Bev AP: 38.7652, 35.5619, 34.8579 3D AP: 17.2034, 15.0750, 15.0462 AOS AP: 36.46, 32.01, 31.09 Car AP_R40@0.70 0.70, 0.70: Bbox AP: 34.1645, 29.5950, 28.4427 Bev AP: 36.4198, 32.3735, 31.4788 3D AP: 14.3970, 12.3787, 11.8986 AOS AP: 33.77, 29.22, 28.09 Car AP@0.50 0.50, 0.50: Bbox AP: 59.9559, 52.9196, 51.8491 Bev AP: 59.0141, 51.9162, 51.0499 3D AP: 49.5782, 42.9737, 42.7284 AOS AP: 59.16, 52.15, 50.94 Car AP_R40@0.50 0.50, 0.50: Bbox AP: 59.3984, 51.7946, 50.4342 Bev AP: 59.0424, 50.9988, 48.9314 3D AP: 48.2886, 41.6285, 39.8300 AOS AP: 58.52, 50.89, 49.54
Well, maybe there are some bugs in my code. I did it a year ago and I forgot many details.
Okay, thank you very much. After I reproduced it, I think there might be some issues with my code, too. But in fact, I almost haven't changed the code, so I don't know how to solve it. I see that you have the same problem with the same performance, so I asked if you have solved it.
hello,so have you solved the dataset split question？
No, with such a small amount of Astyx data, different dataset partitioning should have a significant impact on the results.
If there is no split provided by the author, it may not be possible to directly compare the original results in the paper. I suggest just using the results reproduced by oneself. In fact, I didn't use the Astyx dataset in the end because there are already many larger 4DRadar datasets available now.
Could you please upload the ImageSets file? I just randomly split the astyx dataset and follow your instructions but only get very poor performance.