search

valeoai / RADIal

[CVPR 2022] RADIAl: Raw High-Definition Radar for Multi-Task Learning
180 stars 57 forks source link

Excessive gap in mAR indicators #66

Open haomingHu-dd opened 11 months ago

haomingHu-dd commented 11 months ago

I will pull it locally, configure the running environment, get the model file through training, and then use the evaluation.py file to evaluate. The mAP indicators obtained are all greater than 80% (in hard mode), but only 64% in the paper, whether the code It is caused by different versions, or the data set has changed.

1703084012035

Blume95 commented 1 day ago

I have the same question here. I train the model with the original config file 100 epochs. I have better performance than the report in the paper. 93.23 mAP and 95.14 mAR. Range error: 0.14 m > 0.11 m and Angle error: 0.11 degree < 0.17 degree. I would like to know, something changed in the code or ?