MapTR icon indicating copy to clipboard operation
MapTR copied to clipboard

[ICLR'23 Spotlight & IJCV'24] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction

Results 129 MapTR issues
Sort by recently updated
recently updated
newest added

首先感谢作者的开源! 使用len_queue=2训练时报错显示 File ".../maptr/projects/mmdet3d_plugin/maptr/modules/transformer.py", line 140, in get_bev_features NameError: name 'rotate' is not defined 代码中rotate函数确实没有给出定义,请问作者能将rotate函数开源/补充进项目中吗,谢谢。

https://github.com/hustvl/MapTR/blob/main/projects/mmdet3d_plugin/datasets/map_utils/mean_ap.py#L122 This is the detailed code url

Hi @LegendBC , Thanks for such a wonderful job ! When will the argoverse2 related works be released? (data converter, cfgs, models, training etc...)

谢谢你们的工作!请问网络最后的点列输出结果是在哪个坐标系下的呢? Thank you for your work! Which coordinate system are the final point outputs of the network in?

Thanks for your excellent work! Will the code for lidar and fusion modality be released?

mmdet3d是0.17.0的版本,我提前装好了这个版本,最后编译成功了,但是运行训练命令的时初始化报错了,应该是有numba的版本的问题,numba依赖于mmdet3d的版本,但是感觉不只是这个问题。install.docment里面的环境都比较偏老一点,有没有稳定的一个环境呢?但是我编译成功了啊。可以帮忙解答一下吗?谢谢

I would like to use a larger bev size. I change the bev_h and bev_w in the config file, but got the following error. ` assert (spatial_shapes[:, 0] * spatial_shapes[:,...

Could someone give me some suggestions? Why init the last layer of cls_branches like this? bias_init = bias_init_with_prob(0.01) for m in self.cls_branches: nn.init.constant_(m[-1].bias, bias_init)

I have followed the installation steps in this repo, however, when I ran `python setup.py develop` for mmdetection3, I got the following error: ` Installed /xxx/MapTR/mmdetection3d Processing dependencies for mmdet3d==0.17.2...

Can I train MapTR on multi-node GPU machine?