Xiaopei Wu

The depth map is only generated from the center patch (352x1216) of an RGB image due to the different sizes of RGB images in the KITTI dataset. When we gather...

The model can be trained end to end. For convenience, we use the pre-trained depth completion model to generate pseudo clouds.

Hi, did you train SFD on train + val set before you submit the results to server? Although this way can get higher results, it is not easy to choose...

The recall has a big impact on the performance of SFD on the test set. To get more details about recall, you can print [ret](https://github.com/LittlePey/SFD/blob/03f83de462033e4f840f38f0cead44c506b3f1c5/pcdet/datasets/kitti/kitti_object_eval_python/eval.py#L587) when you evaluate each checkpoint.

We train it on train + val set. In this situation, it is not intuitive to choose a checkpoint with high recall. You may need to visualize the results on...

Hi, we have not tried training SFD with DIoU. We didn't encounter Nan situation when training SFD with GIoU, maybe you need change an environment.

您好，目前还没有避免这种情况的手段。随着训练的进行，roi会更准，问题能得到改善，但没法完全避免

Hi, you can run [this script](https://github.com/LittlePey/SFD/blob/03f83de462033e4f840f38f0cead44c506b3f1c5/pcdet/datasets/kitti/kitti_dataset_sfd.py#L570) to generate info pickle files.

Hi, you can use the following command: ``` python -m pcdet.datasets.kitti.kitti_dataset_sfd create_kitti_infos tools/cfgs/dataset_configs/kitti_dataset_sfd.yaml ``` And it seems that you did not set 'training=False' when initializing KittiDatasetSFD as [us](https://github.com/LittlePey/SFD/blob/03f83de462033e4f840f38f0cead44c506b3f1c5/pcdet/datasets/kitti/kitti_dataset_sfd.py#L523). When training=False,...

Hi, the current version of our code seems only support 1 sample on each GPU, could you show the error information after you train with batch size of 1.