Why pseudo point cloud results in this proposed method seem too far different from LIDAR?

[pengweiweiwei]

In original LIDAR point cloud, there are several cars on the highway. But cant tell these items in this codebase using pretrained models.

[Zong-Ming-Jing]

Do you fix the problem ? I have the same problem too !

[mileyan]

Hi @pengweiweiwei @Zong-Ming-Jing Could you please send a point cloud sample (the .bin file) to [email protected]? I can help you address this question.

[weiyangdaren]

我遇到了和你类似的问题。我使用的是pytorch0.4.1和作者提供的模型finetune_300.tar。 然后我改成pytorch0.4.0，但是发生了错误“num_batches_tracked”。于是我将psmnet/sunmission.py里的 model.load_state_dict(state_dict['state_dict']) 改成 model.load_state_dict({k.replace('.num_batches_tracked', ''):v for k,v in state_dict['state_dict'].items()}) 它成功了 希望能帮到你

[vkorotkine]

I actually had a similar issue. The approach ends up working perfectly fine, but the point cloud results are different. LIDAR (red), Your provided pseudo-lidar point cloud (green), and the one computed from a disparity map generated by psmnet (blue). This is the top view for image 000003 from the object training dataset.