Some questions about eval result

[mzy991215]

Thank you for your open-source code. However, when I trained with your tiny version config file, the results were very poor. Could you help me?
I trained the tiny model on 4 3090 GPUS, but i got the eval result as below:

mAP: 0.0419
mATE: 1.0966
mASE: 0.4038
mAOE: 0.7717
mAVE: 0.9362
mAAE: 0.3531
NDS: 0.1745
Eval time: 162.8s

Per-class results:
Object Class AP ATE ASE AOE AVE AAE
car 0.192 0.892 0.183 0.446 0.934 0.252
truck 0.007 1.113 0.332 0.581 0.802 0.303
bus 0.018 1.092 0.268 0.491 2.418 0.475
trailer 0.006 1.323 0.397 0.872 0.242 0.068
construction_vehicle 0.000 1.000 1.000 1.000 1.000 1.000
pedestrian 0.015 1.073 0.334 1.419 0.859 0.516
motorcycle 0.000 1.086 0.359 1.030 0.939 0.157
bicycle 0.000 1.186 0.419 0.861 0.296 0.054
traffic_cone 0.034 1.102 0.410 nan nan nan
barrier 0.147 1.100 0.336 0.247 nan nan