Yineng Xiong

> > FCOS3D本身不支持，但应该可以通过微调一些地方也加上相应的模块。 bbox2d的结果应该是没有经过nms后处理的，因为本身主任务还是 3D 检测，加这个 branch 除了 regularize 3D 检测框之外只是为了适配 KITTI 上可以单独输出 2D 框的需求来进行评测，不过这个地方确实还可以优化来提升一下 bbox2d 结果。 > > 谢谢回答！单目检测现在都只能原图进行训练，不能改变resize是吧，然后大概多久会支持这个功能啊，如果自己改代码，需要从哪下手和注意哪些问题，谢谢！ Looking forward to `Resize3D` too，`cam2img` should be changed when resizing...

> 1. Resize3D and more pipelines will be supported in a month. > 2. I did not try `RandomShiftScale` in SMOKE, but `Resize3D` may not bring gains on nuScenes from...

Same issue, visualization results are correct on KITTI dataset, and results on nuScenes are questionable, and the problem is not related to the NMS bug

Can batch_size > 1 during training get comparable performance with batch_size=1 in your experiment? thanks If batch_size>1 can get similar performance, it can save lot of time

also segment Fault

So if I want to use model e.g. torchvision, which built with nn.BatchNorm I should: 0. model = torchvision.models.resnet50() 1. model = bnconvert(model) 2. model = DataParallelWithCallback(model) 3. model.cuda() Am...

> Correct. I suspect the reason is the following: > > > > The implementation requires that each module on different devices should invoke the batchnorm for exactly SAME amount...

@eduardinjo I encountered Check failed: work_element_count > 0 (-1072000000 vs. 0) Aborted (core dumped) when I run ops_test.py, did you resolve this problem?

me too, only get about ~18 on Car AP40 using the config in the repo