Hello, I have got confusion with the training label and the detection range.

[kronaldo123]

I dont have KITTI data, so I reproduced the model with offered checkpoint on my own dataset, I found that only "car" objects can be detected. So I wonder if you only train the dataset with only "car" label？

[Ivan-Tang-3D]

Without finetuning on custom dataset, the domain gap between datasets could lead. to the dramatic decrease in the perfromance.

[kronaldo123]

Without finetuning on custom dataset, the domain gap between datasets could lead. to the dramatic decrease in the perfromance.

Thanks, I have noticed the domain gap if i change the dataset. Here, #12 I have found that they only train "car" label.

[kronaldo123]

But I still have a little confusion with the results on KITTI, they can not detect the object on farther objects(about 20 meter?). Have anyone met the same situation？

[Ivan-Tang-3D]

It could be in connection with the training details. The depth range detected could not include the 20 meters.