Liu Yue

pos_inds.numel() == 0 is not allowed in fact

2

### in fcos3d.py: FCOS3DLoss, it writes ```python if pos_inds.numel() == 0: losses = { "loss_box3d_quat": box3d_quat.sum() * 0., "loss_box3d_proj_ctr": box3d_ctr.sum() * 0., "loss_box3d_depth": box3d_depth.sum() * 0., "loss_box3d_size": box3d_size.sum() * 0.,...

**Describe the feature** **Problem** nuscenes_converter.py is too slow, and can be improved. One of the bottlenecks is in data_converter.nuscenes_converter._fill_trainval_infos, line 207 "cam_path, _, cam_intrinsic = nusc.get_sample_data(cam_token)". actually, line207 only want...

add feature / fix bug: I fixed the `kNRows` feature in forward

2

Thank you for sharing this splendid work! I found that `kNRows` is always `1` in `original selective_scan`, and I observed that if I use greater `kNRows` in `selective scan`, the...

## ❓ Questions and Help in METRICS.md, it reads that one symptom of misusing ngR as R is predcls R@100 >75%. is there any theoretical reason behind this?