YangChen1234567
YangChen1234567
> yes, it's different. But I visualized the detection result with test.py, found that the orientation and location is correct. It seems that there is no relationship between low mAP...
> Please ignore my suggestion, it seems I had some mix-up regarding bbox definitions with my own code. > > Any progress on the nuScenes dataset? no progress.....
> @YangChen1234567 Is it only Nuscenes? Or did you also test it on waymo with bad results? I just train DSVT on Nuscenes, didn't train or test it on waymo.....
> I will train it on waymo the next couple of days and report if I am able to reproduce the results. > > However, it would be quite nice...
@MarvinKlemp Hi~ Have you reproduced DSVT on Waymo dataset?
> @YangChen1234567 on waymo I can recreate the results. > > I could try to recreate your NUS training. But I need some time to generate the GT for NUS....
> BTW: I can confirm that with your config I get quite bad results. > > ``` > Evaluating bboxes of pred_instances_3d > mAP: 0.3820 > mATE: 0.3414 > mASE:...
> 我重新适配一下版本号解决了。 之前出问题时安装的版本号 mmcv 2.0.0rc4 mmdet 3.0.0rc5 mmdet3d 1.3.0 mmengine 0.10.1 更新了mmdet与mmengine的版本 mmdet 3.0.0rc5 -> mmdet 3.0.0rc6 mmengine 0.10.1 -> mmengine 0.10.2 hello,请问你复现mmdet里bevfusion的性能了吗？
> 没有，我刚才在issue上回复了，您可以看一下 小手冰凉 ***@***.*** > […](#) > ------------------ 原始邮件 ------------------ 发件人: ***@***.***>; 发送时间: 2024年2月16日(星期五) 下午2:49 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [open-mmlab/mmdetection3d] [Bug] BEVFusion无法训练 (Issue #2792) 我重新适配一下版本号解决了。 之前出问题时安装的版本号 mmcv...
When I ues bevfusion in our own dataset(only one camera too), I get a drop in the fusion model's performance， have you found the reason??