Change the point_cloud_range, can not converge

[M201871109]

When training the model, I use the pandaset dataset, and I change the point_cloud_range from X∈[-51.2, 51.2] to X∈[-100, 100], then results show that it is hard to converge, even predict in the training set, only a part of the object will be recalled. And I tried to train the model in a very small training set(only 10 samples), and there is no bbox prediction in training set. So why? Maybe the cls_loss （focal_loss）funciton's parameters should be adjusted?

[JiyuanWANG12138]

When training the model, I use the pandaset dataset, and I change the point_cloud_range from X∈[-51.2, 51.2] to X∈[-100, 100], then results show that it is hard to converge, even predict in the training set, only a part of the object will be recalled. And I tried to train the model in a very small training set(only 10 samples), and there is no bbox prediction in training set. So why? Maybe the cls_loss （focal_loss）funciton's parameters should be adjusted?

Hello, do you know how to solve this problem now?