The size and test result of trained model.

[ShiAngWang]

你们好，十分感谢你们做出的工作，我有一个关于训练出的模型大小的问题。我没有修改代码，直接训练200epoch出的模型大小为 225M，而您发布在google drive 上的模型只有 75M，但是两个模型的参数量是相同的，都是19.46M，所以想请问下您是做了什么尺寸上的优化吗？

另外想问下评测结果，我用您google drive 上的模型做test，发现评测结果和paper中并不完全一致，例如 Kitti Car AP 3D（0.70 0.70 0.70） 这个，我跑出的结果为 26.3332, 19.0269, 15.9994，而paper 记录为 22.50 16.46 13.95；而pedestrain AP 3D(0.50,0.25, 0.25) 我跑出来的test结果为10.0956, 9.5456, 7.7280，则不如paper中的13.10 8.41 6.94，请问paper中使用的是不是不是 google drive上的模型呀？

另外，我自己训练的模型，pedestrain AP 3D(0.50,0.25, 0.25) 结果达到了 29.7218, 24.0618, 19.0713，远远超过了现有结果，也是十分有趣
期盼您的回复

Hello, thank you very much for your work. I have a question about the size of the trained model. I didn't modify the code. The model size from the direct training of 200 epochs is 225M, while the model you published on Google Drive is only 75M. However, the parameters of the two models are the same, which is 19.46M. So I want to ask what size optimization you have made?

In addition, I would like to ask about the test results. I used the model on your Google drive to test, and found that the evaluation results are not completely consistent with those in the paper. For example, ** Kitti Car AP 3D (0.70 0.70 0.70) **, the results I ran were 26.3332, 19.0269, 15.9994, while the paper record was 22.50 16.46 13.95; ** pedestrian AP 3D (0.50, 0.50, 0.50) ** The test results I run are 10.0956, 9.5456, 7.7280, which is not as good as 13.10 8.41 6.94 in the paper. Is the model used in the paper on the Google drive?

By the way, the results of my own training model, ** pedestrian AP 3D (0.50, 0.50, 0.50) ** have reached 29.7218, 24.0618, 19.0713, far exceeding the existing results, which is also very interesting

Looking forward to your reply

[Lpy15540250240]

bro，share you enviorment .@ShiAngWang