cnzzx / gemap Goto Github PK

Hi, have you tried setting the pred_z_flag and gt_z_flag within z_cfg to True to train and go for evaluation of accuracy? I tried pred_z_flag and gt_z_flag within z_cfg set to True but reported an error, now I'm trying to modify the associated loss for the added z values. If you haven't done the above attempts, can you give me some advice? Really thank you very much!!!

error in multiscale_kernel_attn_forward_cuda: invalid configuration argument
error in multiscale_kernel_attn_backward_cuda: invalid configuration argument
Hi, when I try to train my own data, the above error is reported, can I ask if you guys have encountered similar errors or have any recommended solution ideas?

Hello, really impressive work!

I was wondering whether you compute BEV lanes' poins in real world/lidar coordinates for nuscenes.
Because i would integrate your method with some 3d detection models, and I would like to make sure to have exactly the same coordinate systems (or at least to know which transformation I need to get it).

Thank you very much! :)

when i run python tools/gemap/custom_av2_map_converter.py --data-root ./data/hdmap/av2/sensor/ --out-dir ./data/hdmap/av2/sensor/full
an error message appears:
networkx.exception.NodeNotFound: target node [(1599.82, 226.695, 14.85), (1622.585, 225.45, 14.965), (1648.38, 274.055, 13.135), (1602.89, 225.295, 14.935), (1577.315, 256.68, 14.1), (1576.23, 259.925, 14.05), (1645.89, 276.635, 13.215), (1650.755, 271.665, 13.11), (1625.335, 230.325, 14.91)] not in graph
Can you help me with this one?

Hi, thanks for your great work.

After reading your paper and code, I feel confuse about the attn_mask definition. According to MultiheadAttention.attn_mask, 'For a binary mask, a True value indicates that the corresponding position is not allowed to attend.'

Hi author, thanks for the great work.
Got a few questions:

1. How long does it take to do full 110 epochs with 8 3090Ti？
2. What is the performance for V2-99 backbone when training for 24 epochs?

Hello, have you tried expanding the range even further? For example, a total of 100 meters in front and behind, 120 meters, or even farther.