Jiazhen Li

I know that some people uesd detectreon2-v0.2.1 before,but there was no amp function.

wrapper最后一行改了 ---Original--- From: ***@***.***> Date: Tue, Jun 4, 2024 17:08 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [ChenhongyiYang/QueryDet-PyTorch] attribute:'torch.device' objecthas no attribute 'device' (Issue #90) 请问解决了吗，我也是这个问题 — Reply to this...

训练时长跟你的参数设置有关，跟这个地方貌似关系不大。 ---Original--- From: ***@***.***> Date: Tue, Jun 11, 2024 10:45 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [ChenhongyiYang/QueryDet-PyTorch] attribute:'torch.device' objecthas no attribute 'device' (Issue #90) 我改了最后一行之后的训练时长就很长了，请问有更好的办法吗 — Reply to this...

> 嗨，SpConv作者，感谢您的出色工作！我的spconv版本是spconv-cu113 2.3.6。我的 cuda 版本是 cu11.3 当我测试我的模型时，但安装 spconv 后总是出现导入错误： 回溯（最近一次调用）：文件“tools/create_data.py”，第 13 行，从 det3d.datasets.utils.create_gt_database 导入 create_groundtruth_database 文件“/media/wxr/work/lidar_detection/CenterPoint/CenterPoint/Det3D/det3d/datasets/**init**.py”，第 4 行，从 .kitti 导入 KittiDataset文件“/media/wxr/work/lidar_detection/CenterPoint/CenterPoint/Det3D/det3d/datasets/kitti/**init**.py”，第 1 行，从 .kitti 导入 KittiDataset 文件“/media/wxr/work/lidar_detection/CenterPoint/CenterPoint/Det3D/det3d/datasets/kitti/kitti.py”，第 7 行，从 det3d.core...


I made the same mistake，It seems that the version of spconv will be lowered.

Hello, do you have another problem with inconsistent tensor sizes

I changed this problem for a long time yesterday, and found that it could not be solved. I inferred that it was not the comment problem, but the spconv version...

> > 您好，您是否还有其他张量大小不一致的问题 > > 是的，问题是“RuntimeError：张量的扩展大小必须与非单例维度的现有大小匹配，目标维度为 [256,3,3,256]，但张量维度为 [3,3,256,256]” 我解决了这个问题，我把2.3.1.0的顺序改成了0.2.3.1后就没有错误了，但是我跑的结果很差，coco的结果只有0.几