Hello, I tried to build and train your code, M3DSSD. But I have one problem with training.

I ran the following command with the guide you provided. python3 scripts/train_rpn_3d.py --config=kitti_3d_base --exp_name base

But I got the following error.

  File "scripts/train_rpn_3d.py", line 324, in <module>
    main(args)
  File "scripts/train_rpn_3d.py", line 114, in main
    dataset = Kitti_Dataset_torch(conf, paths)
  File "/workspace/lib/dataloader.py", line 892, in __init__
    generate_anchors(conf, self.imdb, paths.data_cache) # generate anchors to conf.anchors
  File "/workspace/lib/rpn_util.py", line 118, in generate_anchors
    ols = iou(anchors[:, 0:4], normalized_gts[:, 0:4])
IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

I checked that normalized_gts.shape is (0,), so I think there is no loaded data in normalized_gts. Then I took a look at the rpn_util.py, dataloader.py, and train_rpn_3d.py. I found out that normalized_gts was from "imdb", and "imdb" was from train_imdb.pkl" file.

But I don't have the train_imdb.pkl file, so any data cannot be loaded to imdb and normalized_gts.

How can I get the pkl file and solve this problem?

Thank you!

Thank you very much for your excellent work. I have a question, did you only use PyTorch ==0.4.1 for conditional configuration, and what are the other environmental conditions? 作者您好，非常感谢您的出色工作。我有一个问题，在条件配置方面，你只用到了pytorch==0.4.1吗，别的环境条件是什么呢？

Hello, author, for the test written in the readme, is this evaluating the model on the validation set, or is it generating prediction labels on the test set?

code:Modify the conf_path and weights_path to run test. python scripts/test_rpn_3d.py

Hello, author, I have a question. Among the evaluation results obtained from the training, there are two indicators of the car: AP@R11 and AP@R40. Are these two results on the val datasets? Or AP@R40 is the result on the testing datasets? 作者您好，我有个问题。在训练得到的评估结果中，有汽车在AP@R11和AP@R40两种指标，这两种都是在验证集上的结果吗，还是AP@R40是测试集上的结果 ? bbox AP:93.4128, 86.0374, 76.3270 bev AP:28.6008, 24.1458, 19.8311 3d AP:23.5013, 18.5377, 16.8939 aos AP:93.18, 85.31, 75.28 Car [email protected], 0.70, 0.70: bbox AP:94.4730, 89.1243, 75.0301 bev AP:25.3284, 19.6154, 15.5039 3d AP:17.6740, 13.5385, 10.9201 aos AP:94.25, 88.29, 74.02 Pedestrian [email protected], 0.50, 0.50: bbox AP:72.4768, 64.8912, 56.7276 bev AP:8.2554, 6.1055, 6.1303 3d AP:7.4127, 5.7886, 5.7870 aos AP:56.16, 48.96, 42.46 Pedestrian [email protected], 0.50, 0.50: bbox AP:73.0240, 62.6571, 53.7581 bev AP:6.3362, 5.0528, 4.4319 3d AP:5.5379, 4.2182, 3.5477 aos AP:56.40, 47.08, 40.11 Cyclist [email protected], 0.50, 0.50: bbox AP:71.7324, 55.8072, 48.8942 bev AP:7.4572, 5.5556, 5.4545 3d AP:6.1668, 5.3124, 5.3270 aos AP:62.99, 47.76, 42.04 Cyclist [email protected], 0.50, 0.50: bbox AP:70.4419, 52.3072, 50.2627 bev AP:3.8131, 2.2562, 1.9963 3d AP:2.7601, 1.5694, 1.5516 aos AP:61.29, 44.23, 42.05

Second question:I load the conf file and weights from the results of the training into the test file： conf_path = '/ conf_path data01 / zq/M3DSSD - master/output/base / 20211215 _161854_14. 29574471493946 / kitti_3d_base. Conf' weights_path = '/ weights_path data01 / zq/M3DSSD - master/output/base / 20211215 _161854_14. 29574471493946 / weights/model_dla102_best_pkl' When I run Python scripts/test_rpn_3d.py ,The following error has occurred. What is the problem?

(zq5) omnisky@node01:/data01/zq/M3DSSD-master$ CUDA_VISIBLE_DEVICES=5 python3 scripts/test_rpn_3d.py Traceback (most recent call last): File "scripts/test_rpn_3d.py", line 17, in < module> From lib.dataloader import Kitti_Dataset_torch, Kitti_Dataset_dali ImportError: cannot import name 'Kitti_Dataset_dali'

Dear author, Thank you very much for your outstanding contribution. Now I have this big problem. Because you did not provide the pre-training model, I could not compare with your simulation results. This is really important to me. I would appreciate it if you could publish your pre-training model. 尊敬的作者您好： 非常感谢您做出的突出贡献。现在我面临这一个很大的问题。因为您没有给出预训练的模型，所以我没有办法和您的仿真结果进行比较。这对我来说真的很重要。希望您能发布您的预训练模型的话，我将不胜感激。

Thank you for the great work! I have a question about the content of the paper. In 3.4. 2D-3D Prediction and Loss. Output transformation Why should parameters of 2D/3D box be recovered from detection outputs? And how to explain the element-wise product in the formula? Maybe my understanding is not in place. I'm looking forward to your answer, thanks :).