search

hustvl / MapTR

[ICLR'23 Spotlight & IJCV'24] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction
MIT License
1.08k stars 163 forks source link

Accuracry is significantly reduced when extending point cloud range #54

Open quangnhat185 opened 1 year ago

quangnhat185 commented 1 year ago

Hi,

I tried to reproduce your result using the point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0] which you provided in the config file and it works wells. However when I extended the range to point_cloud_range = [-50.0, -50.0, -2.0, 50.0, 50.0, 2.0] and train the model again, the prediction result is significant worse.

I expect that it would probably maintain a high confidence score in the previous range, but it was no the case. Could you kindly share your opinion on this? Do you have any advice if I want to extend the point_cloud_range?

ailib commented 1 year ago

Can you share the environment configuration， I have been configuring it for a long time, but it all fails，thanks

ailib commented 1 year ago

requirement.txt && the conda version, best!

Rkyzzy commented 1 year ago

Same here, when extending the pointcloud range while remaining the hyperparameters the same, the map drops from 50 to ~30, Any idea on the direction to tune the hyperparamter in this case? my pointcloud range is around the same as @quangnhat185 . Any idea will be helpful, thanks!