search

Cc-Hy / CMKD

Cross-Modality Knowledge Distillation Network for Monocular 3D Object Detection (ECCV 2022 Oral)
Apache License 2.0
107 stars 9 forks source link

AttributeError: 'EasyDict' object has no attribute 'POST_PROCESSING' #25

Closed xiaoxusanheyi closed 1 year ago

xiaoxusanheyi commented 1 year ago

AttributeError: 'EasyDict' object has no attribute 'POST_PROCESSING' image

执行命令python train_cmkd.py --cfg /home/ubuntu/CMKD/tools/cfgs/kitti_models/CMKD/CMKD-scd/cmkd_kitti_eigen_R50_scd_bev.yaml，训练30个epoch出现上面那个错误，也就是你们建议的训练技巧，是怎么回事？

Cc-Hy commented 1 year ago

That's ok, don't care about it. After training it will automatically run inference and the xxx_bev.yaml config file is not complete so it will have a problem with inference. And when you complete detection training the error will disappear.

xiaoxusanheyi commented 1 year ago

也就是说这条命令python train_cmkd.py --cfg /home/ubuntu/CMKD/tools/cfgs/kitti_models/CMKD/CMKD-scd/cmkd_kitti_eigen_R50_scd_bev.yaml是正确的，然后训练完去执行下一条命令python train_cmkd.py --cfg /tools/cfgs/kitti_models/CMKD/CMKD-scd/cmkd_kitti_eigen_R50_scd_V2.yaml --pretrained_img_model +上面命令训练完的第30个epoch 训练10个后测试吗？

Cc-Hy commented 1 year ago

Yes, after you run the second detection training command, this error will not occur.

xiaoxusanheyi commented 1 year ago

是的，你运行第二次检测训练命令后，就不会出现这个错误了。

1）最早的时候，我也是那样训练的，但是最后检测的出的结果，不管是汽车、行人还是自行车的bev、3d数据都是00，00，稍后我会在训练，测试一遍，看看是否数据是否正常。 2）在多GPU训练的命令中有加载的--pretrained_lidar_model checkpoints/second_teacher.pth，但为什么在We recommand to use the following training skill (BEV first, then whole model)命令训练方式中就没有了呢？ 上述1）的问题是不是也是此原因呢？

Cc-Hy commented 1 year ago

In 'We recommand to use the following training skill' part, the commands are simplified to explain the training skill. And you should complete the commands with the rest arguments.

For all 0 results, you may upload the log file and I can look into it to find the potential problems.

TimGor1997 commented 7 months ago

是的，你运行第二次检测训练命令后，就不会出现这个错误了。

1）最早的时候，我也是那样训练的，但是最后检测的出的结果，不管是汽车、行人还是自行车的bev、3d数据都是00，00，稍后我会在训练，测试一遍，看看是否数据是否正常。 2）在多GPU训练的命令中有加载的--pretrained_lidar_model checkpoints/second_teacher.pth，但为什么在We recommand to use the following training skill (BEV first, then whole model)命令训练方式中就没有了呢？ 上述1）的问题是不是也是此原因呢？

大佬您好！我想问一下，最终00，00的问题解决了吗？请问是什么原因造成的呢？